Full Audio at

🚀 Welcome to a Special Double-Feature Deep Dive on AI Unraveled.

Today, we expose the two “invisible” economies reshaping business in 2026. Externally, we are entering the era of Business-to-Agent (B2A), where your next customer is a piece of code that doesn’t care about your brand. Internally, we are witnessing the explosion of Shadow AI, where 98% of enterprises are running on “underground” intelligence that IT can’t see or control.

We dissect the rise of “Moltbot” and the “Invisible Shelf,” explain why the CMO is being replaced by the CTO, and reveal how employees are building a parallel infrastructure to bypass corporate bureaucracy.

Strategic Pillars & Key Topics:

🤖 Part 1: The Rise of B2A (Marketing to Machines)

• The “Ruthless Arbiter”: Why autonomous agents like Moltbot and Gemini prioritize API latency and JSON data over brand storytelling.

• The Invisible Shelf: In a world of AI buying, there is no “Page 2.” You are either the algorithmic answer, or you do not exist.

• New Protocols: The shift from SEO to AIO (Agent Optimization), and the rise of the Universal Commerce Protocol (UCP) as the “common language” of shopping bots.

• CMO vs. CTO: Why marketing is no longer about persuasion, but about data ontology, making the CTO the new custodian of the brand.

🕵️ Part 2: The Shadow AI Economy (The Internal Rebellion)

• The 95% Failure Rate: While corporate AI pilots stall in “purgatory,” the Shadow AI economy is flourishing, with 78% of employees bringing their own tools (BYOAI) to work.

• The “Local” Underground: How employees are using tools like Ollama and LM Studio to run powerful models locally, bypassing corporate firewalls and DLP systems.

• The GenAI Divide: The gap between executives (who buy “brittle” enterprise tools) and employees (who use agile, consumer tools) has created a massive security blind spot.

Credits: This podcast is created and produced by Etienne Noumen, Senior Software Engineer and passionate Soccer dad from Canada.

Keywords

Business-to-Agent, B2A, Shadow AI, Machine Customers, Moltbot, Universal Commerce Protocol, UCP, Agent Optimization, AIO, BYOAI, Shadow IT, Ollama, Local Inference, Invisible Shelf

🚀 Reach the Architects of the AI Revolution

Want to reach 60,000+ Enterprise Architects and C-Suite leaders? Download our 2026 Media Kit and see how we simulate your product for the technical buyer: https://djamgamind.com/ai

Connect with the host Etienne Noumen, Senior Software Engineer and passionate Soccer dad from Canada

LinkedIn: https://www.linkedin.com/in/enoumen/

Youtube: https://youtube.com/@enoumen

X: https://twitter.com/enoumen

Email: etienne_noumen@djamgamind.com

🛑 Stop fearing Shadow AI. Orchestrate it.

Your employees are using AI whether you approve it or not. The problem isn’t the innovation; it’s the lack of visibility.

Airia is the “Control Plane” for your entire enterprise AI stack. It allows you to say “Yes” to innovation without compromising security or blowing up your API budget.

Why we are partnering with them:

• Unified Security: Get automated threat detection and governance across all your models and agents.

• Real-Time Cost Control: Stop guessing your token usage. Manage budgets and quotas from a single dashboard.

• Model Agnostic: Whether your devs want low-code or pro-code tools, they can prototype safely in a production-like environment.

👉 Secure Your Stack: Get the Airia Demo

1️⃣ From Shadow IT → Shadow Intelligence

Shadow IT isn’t new.

Employees have always adopted unsanctioned tools:

• Dropbox before corporate storage

• WhatsApp before internal chat

• Google Docs before Office 365 rollout

But Shadow AI is categorically different.

Because employees aren’t just adopting tools — they’re deploying intelligence layers:

• Browser copilots writing emails and reports

• Personal LLM subscriptions analyzing corporate data

• Local open-source models running on laptops

• AI coding agents committing production code

• Fine-tuned workflows automating decisions

This isn’t unsanctioned software.

It’s unsanctioned cognition.

2️⃣ How Big Is the Shadow AI Economy?

Short answer: Massive — and underreported.

Enterprise surveys across 2025 show a widening gap between official adoption and employee reality:

• A majority of knowledge workers report using generative AI at work — even when policies are unclear or restrictive.

• In multiple enterprise studies, employee usage outpaces sanctioned deployment by 2–3×.

• Security teams consistently rank unsanctioned AI usage among their fastest-growing visibility gaps.

What’s happening is simple:

Executives approve AI top-down.
Employees deploy it bottom-up — faster.

In Fortune 500 environments, this creates three parallel AI layers:

LayerVisibilityControlOfficial copilotsHighHighTeam-level toolsMediumMediumPersonal AI stacksNear zeroNone

The third layer — invisible but productive — is the Shadow AI Economy.

3️⃣ What’s Driving Underground Adoption?

⚡ Productivity Pressure

Employees aren’t using AI to rebel.
They’re using it to keep up.

If one analyst uses AI to produce reports 5× faster, others must follow — policy or not.

AI becomes a competitive survival tool at the individual level.

🌐 Frictionless Access

No VPN required.
No installation required.
No approval required.

A credit card + browser = enterprise-grade intelligence.

This accessibility collapses traditional IT gatekeeping.

🧠 Capability Asymmetry

Employees often discover practical use cases faster than leadership:

• Sales reps generating prospect briefs

• Marketers synthesizing campaign data

• Developers debugging legacy code

• Finance teams modeling scenarios

Innovation is happening at the edge — not the center.

4️⃣ Are Security Teams Losing Visibility?

Yes — and faster than policies can keep up.

Shadow AI introduces three acute risk vectors:

🔓 Data Leakage

Employees paste sensitive data into external models:

• Contracts

• Financials

• Source code

• Customer data

Even when models claim not to train on inputs, governance teams lose audit trails.

🧬 Model Contamination

Unofficial outputs enter official systems:

• AI-generated code in production

• Synthetic research in reports

• Fabricated citations in strategy decks

This creates “silent integrity drift.”

🕵️ Workflow Obfuscation

Decisions appear human-made — but are AI-assisted.

This breaks traceability:

• Who made the decision?

• What data informed it?

• Which model influenced it?

Accountability blurs.

5️⃣ The Policy Lag Problem

Security teams face an impossible equation:

AI adoption speed > Policy creation speed

By the time governance frameworks are written:

• New tools exist

• New models launch

• New use cases emerge

Policy becomes reactive — never preventative.

This creates what CISOs now call:

“Visibility debt.”

The longer shadow usage persists,
the harder it becomes to map.

6️⃣ Ban vs Legalize — The Strategic Fork

Organizations now face a defining decision.

❌ Path 1 — Prohibition

Ban unsanctioned AI tools.

Pros

• Signals seriousness on compliance

• Reduces immediate data exposure

• Simplifies legal posture

Cons

• Impossible to enforce

• Drives usage deeper underground

• Punishes productivity

• Creates employee resentment

This mirrors early cloud bans — widely ignored.

✅ Path 2 — Managed Legalization

Acknowledge reality and standardize it.

Key moves:

• Approved tool marketplaces

• Secure enterprise wrappers

• Data-loss prevention layers

• Usage telemetry dashboards

• Prompt logging and audits

Instead of banning intelligence…
organizations sandbox it.

7️⃣ Is Banning AI Like Banning Google in 2005?

The analogy is striking — and instructive.

In the early 2000s:

• Some firms banned Google searches

• Others restricted Wikipedia

• Many blocked cloud storage

The result?

Employees used personal devices.

Productivity tools always win against policy friction.

AI is following the same curve — but faster.

Because unlike Google:

AI doesn’t just retrieve knowledge.
It produces it.

Banning AI today is less like banning search…
and more like banning spreadsheets in the 1990s.

Technically possible.
Operationally absurd.

8️⃣ The Executive Strategy vs Employee Reality Gap

Here’s the most important insight:

There are two AI transformations happening simultaneously.

🏢 Executive AI

• Roadmaps

• Procurement cycles

• Governance frameworks

• Multi-year rollouts

Measured. Controlled. Documented.

🧑‍💻 Employee AI

• Browser copilots

• Prompt libraries

• Personal workflows

• Automation scripts

Fast. Invisible. Improvised.

The gap between these two layers is the Shadow AI Economy.

And it’s where most productivity gains are already happening.

9️⃣ What Smart Enterprises Are Doing Now

Forward-leaning organizations are shifting posture:

Instead of asking:

“How do we stop shadow AI?”

They ask:

“How do we make it safe to surface?”

Leading practices include:

• “Bring Your Own AI” governance frameworks

• Enterprise prompt security training

• Internal model hosting

• Secure copilots with audit trails

• AI usage disclosure norms

The goal isn’t suppression.

It’s illumination.

🔮 Strategic Outlook

Shadow AI is not a phase.

It’s a structural feature of the AI era.

Because intelligence is now:

• Cheap

• Accessible

• Portable

• Individually deployable

Which means the enterprise no longer monopolizes capability.

Employees do.

Closing Frame

The Shadow AI Economy exposes a truth executives can’t ignore:

The most important AI systems inside your company
may not be the ones you bought…

…but the ones your employees built without asking.

And the organizations that win won’t be the ones that ban this reality.

They’ll be the ones that learn how to govern it —
without suffocating it.

Full Audio here

🚀 Welcome to AI Unraveled (February 5th, 2026): Your strategic briefing on the business, technology, and policy reshaping artificial intelligence.

Today, we analyze the $300 billion software stock crash triggered by Anthropic’s new legal tools, which are decimating the value of legacy SaaS companies. We also break down Microsoft Copilot’s struggle to retain users, Anthropic’s bold “No Ads” Super Bowl campaign, and a disturbing study showing that “power-seeking” agents are taking over the viral Moltbook platform.

Key Topics:

📉 Markets & The “Selfware” Crisis

• The $300B Selloff: Software stocks (LegalZoom, Thomson Reuters) plunge as investors realize AI agents like Claude Cowork can replace traditional SaaS. Private equity firms exposed to software are also taking a hit.

• Microsoft Copilot Crisis: User preference for Copilot has dropped from 18.8% to 11.5% as confusion and interoperability issues plague the product. Meanwhile, Google’s Gemini is gaining ground.+1

⚔️ The Model Wars

• Anthropic’s “No Ads” Pledge: Anthropic confirms Claude will remain ad-free and launches a Super Bowl ad campaign directly attacking ChatGPT’s new ad tier.

• KLING 3.0: A major upgrade to the video model brings multi-shot consistency, native audio, and 15-second generations, pushing AI video closer to cinema quality.

• Intel GPUs: Intel CEO Lip-Bu Tan announces the company will start making GPUs to challenge Nvidia’s dominance.

🤖 Agents & Security

• Power-Seeking Agents: A new study on Moltbook reveals that 0.52% of agents (”power-seekers”) control 64% of the platform’s engagement.

• OpenClaw Flaw: Security experts warn that viral agents like OpenClaw are vulnerable to “polite hacking”—where attackers simply ask the helpful AI to hand over credentials.

• Google “Screen Automation”: New beta code reveals Gemini will soon be able to tap and scroll your phone to book rides and order food—but users are liable for its mistakes.

🧬 Health & Startups

• Resolve AI Unicorn: AI-for-SRE startup Resolve raises $125M at a $1B valuation to fix software bugs autonomously.

• Luffu: Fitbit founders launch an “intelligent family care” system to monitor health data across devices.

Keywords:

SaaS Crash, Microsoft Copilot Crisis, Anthropic Super Bowl Ad, KLING 3.0, Intel GPUs, Moltbook Power Seeking Agents, Google Screen Automation, Resolve AI Unicorn, Luffu Health, OpenClaw Security.

🚀 Reach the Architects of the AI Revolution

Want to reach 60,000+ Enterprise Architects and C-Suite leaders? Download our 2026 Media Kit and see how we simulate your product for the technical buyer: https://djamgamind.com/ai

Connect with the host Etienne Noumen, Senior Software Engineer and passionate Soccer dad from Canada

LinkedIn: https://www.linkedin.com/in/enoumen/

Youtube: https://youtube.com/@enoumen

X: https://twitter.com/enoumen

Email: etienne_noumen@djamgamind.com

The AI disruption in software is here

• Fears that new AI tools will replace traditional software wiped around $300 billion from the market value of software and data stocks on Tuesday, hitting companies across legal, finance, and tech sectors.

• The selloff began after Anthropic announced new legal tools for its Cowork assistant, sending shares of Thomson Reuters, Legalzoom.com, and London Stock Exchange down more than 12% each.

• Private-funds firms like Ares, KKR, and Blue Owl also dropped sharply because they invested heavily in software companies over the past decade, leaving them exposed to AI disruption risk.

Microsoft Copilot is running into big problems

• Microsoft’s Copilot chatbot, a key part of CEO Satya Nadella’s plan to make the company AI-first, is struggling with confusing brand positioning, interoperability problems, and falling user preference compared to ChatGPT and Google’s Gemini.

• Survey data from Recon Analytics shows that the share of Copilot subscribers who use it as their primary option dropped from 18.8% to 11.5% between July and late January, while Google’s Gemini rose from 12.8% to 15.7%.

• Microsoft has sold only 15 million corporate Copilot seats out of 450 million paid 365 seats, and some companies use just 10% of the Copilot subscriptions they pay for, according to Citi Research analysts.

Anthropic says ‘Claude will remain ad-free,’ unlike ChatGPT

• Anthropic announced that its AI assistant Claude will stay ad-free, setting it apart from competitors like ChatGPT, because the company believes advertising would conflict with Claude’s role as a helpful thinking tool.

• The company argues that AI conversations are different from search engines because users share sensitive or personal information, and ads could create pressure to steer responses toward transactions rather than genuine help.

• Anthropic plans to fund Claude through enterprise contracts and paid subscriptions instead, while exploring lower-cost tiers and features like agentic commerce where Claude handles purchases only when users ask.

The Claude maker has firmly rejected the notion of putting ads in its AI outputs, announcing in a new blog post that hosting ads within chatbot discussions “would be incompatible” with their overall mission of providing “a genuinely helpful assistant for work and for deep thinking.” The post doesn’t mention rivals OpenAI by name, but of course it comes just a few weeks after The House of Altman started inserting ads into its free or low-cost ChatGPT tiers. Anthropic’s latest ad campaign — set to debut on Sunday during Super Bowl LX — is even more pointed. The tagline: “Ads are coming to AI. But not to Claude.”

Intel will start making GPUs

• Intel CEO Lip-Bu Tan announced at the Cisco AI Summit that the company will start producing graphics processing units (GPUs), chips used for gaming and training artificial intelligence models.

• The project will be led by Kevork Kechichian, executive vice president and general manager of Intel’s data center group, with Eric Demers, a former Qualcomm senior vice president, also hired for the effort.

• The initiative appears to be in early stages, with Tan saying Intel plans to develop its strategy around customer demands, marking a notable expansion from the company’s traditional focus on CPUs.

KLING 3.0 is here:

What’s new:

• Multi-shot sequences – The model generates connected shots with spatial continuity. A character moving through a scene maintains consistency across multiple camera angles.

• Advanced camera work – Macro close-ups with dynamic movement. The camera tracks subjects smoothly while maintaining focus and depth.

• Native audio generation – Synchronized sound, including dialogue with lip-sync and spatial audio that matches the visual environment.

• Extended duration – Up to 15 seconds of continuous generation while maintaining visual consistency.

AWS CEO says orbital data centers ‘pretty far’ from reality

• AWS CEO Matt Garman said that data centers in space are “pretty far” from becoming real, despite startups and Amazon founder Jeff Bezos exploring the idea through his rocket company Blue Origin.

• The growth of artificial intelligence demands huge amounts of computing power and cooling, which strains land-based data centers and has pushed cloud companies to consider alternatives like sending equipment to orbit.

• Garman explained that the cost of launching payloads into space is massive, and there are not enough rockets to send the million satellites needed, making the concept not economical right now.

Gemini ‘screen automation’ will place orders, book rides for you

• Google is developing a Gemini feature called “screen automation” that will let the AI place orders and book rides on your phone by scrolling and tapping on your behalf.

• New code in the Google app beta reveals that users must closely monitor what Gemini does because the AI can make mistakes, and users will be held responsible for its actions.

• Google warns that trained reviewers will see screenshots from your device if Keep Activity is on, so users should avoid entering sensitive information or payment details during automation.

Resolve reaches unicorn status:

The startup emerged from stealth in late 2024 with an AI solution for quickly fixing software issues and bugs in commercial and customer-facing systems, where even a few minutes of downtime can cost an enterprise big money. To date, they’ve landed around 20 major customers, including Salesforce, Coinbase, and DoorDash. The company raised a new $125 million round, pushing the valuation up and above $1 billion. They’re not included in the TWiST 500 yet but… hey, it could happen…

The one BIG FLAW in OpenClaw:

OpenClaw and other agents continue to rely on LLMs like Claude, GPT, and Gemini, which represents a major and ongoing security flaw. Because those models are designed to please and accommodate users, while minimizing friction, OpenClaw remains extremely vulnerable to attackers, who exploit this same helpfulness to gain access to users’ private systems and data. ZioSec’s Aaron Walls and Andrius Useckas warn that the next generation of AI security breaches will not rely on sophisticated or cutting edge techniques, but simply hackers asking politely for an agent’s help.

Power Seeking Agents Are Winning on MoltBook!

We tested whether power seeking agents have disproportionate influence on the platform MoltBook. And they do.

• Posts we flagged as power seeking get ~1.5x more upvotes and ~2x more comments than unflagged posts.

• Agents we flagged making these posts have ~2x higher karma and 1.6x more followers than unflagged agents.

• These 65 agents – just 0.52% of all agents on Moltbook – have recieved 64% of all platform upvotes.

Everything Else in AI today

Google’s Gemini app has surpassed 750M monthly active users.

NVIDIA AI Release VibeTensor: An AI Generated Deep Learning Runtime Built End to End by Coding Agents Programmatically.

Open-source AI tool beats giant LLMs in literature reviews — and gets citations right.

In bobsled and speedskating, advanced AI offers Team USA an edge.

Microsoft is building content licensing marketplace for AI

Agentic coding comes to Apple’s Xcode with agents from Anthropic and OpenAI

Nvidia AI chip sales to China stalled by US security review

A new platform lets AI agents pay humans to do the real-world work they can’t

OpenAI accuses Musk’s xAI of systematic evidence destruction in antitrust case

🚀 Welcome to AI Unraveled (February 4th, 2026): Your strategic briefing on the business, technology, and policy reshaping artificial intelligence.

Today, we cover the historic acquisition of xAI by SpaceX, creating the world’s most valuable private company at $1.25 trillion. We also break down OpenAI’s new Codex desktop app for managing agent swarms, and a major medical study showing AI screening detects 27% more aggressive breast cancers.

Full Audio at

Key Topics:

💰 Big Tech Consolidation

• The $1.25 Trillion Merger: SpaceX has officially purchased Elon Musk’s xAI. The combined entity aims to build space-based data centers to bypass Earth’s energy constraints.

• OpenAI’s Strategic Pivot: Reports confirm OpenAI is shifting resources away from long-term research (like Sora/DALL-E) to focus entirely on winning the chatbot war with ChatGPT, leading to internal departures.

🤖 Agents & Dev Tools

• OpenAI Codex App: OpenAI launches a macOS “command center” that lets developers run multiple coding agents in parallel, automating tasks like game development and QA testing.

• Firefox AI Kill Switch: Mozilla announces that Firefox 148 will include a single toggle to block all generative AI features.

• Moltbook Dismissed: Sam Altman calls the viral AI social network “Moltbook” a likely fad, though he validates the underlying agent technology.

🧬 Healthcare & Science

• AI vs. Cancer: A massive Swedish study finds AI screening detects 27% more aggressive breast cancers than standard methods and cuts radiologist workload by 44%.

• Luffu Health: Fitbit founders launch “Luffu,” a new AI system for family caregiving and health monitoring.

⚖️ Law & Geopolitics

• X Raided in Paris: French police have raided X’s offices and issued summons for Elon Musk and Linda Yaccarino regarding algorithmic transparency and Grok-generated content.

🎨 Creative & Design

• Higgsfield Vibe-Motion: A new motion design tool powered by Claude that uses reasoning logic to create editable animations.

• xAI Grok Imagine 1.0: xAI releases an upgraded video model with improved audio and 10-second generations.

Keywords: SpaceX xAI Merger, OpenAI Codex, AI Breast Cancer Screening, Firefox AI Kill Switch, Musk Paris Raid, Higgsfield Vibe-Motion, Luffu Health, Skyryse Helicopters, Grok Imagine 1.0, Sam Altman Moltbook

🚀 Reach the Architects of the AI Revolution

Want to reach 60,000+ Enterprise Architects and C-Suite leaders? Download our 2026 Media Kit and see how we simulate your product for the technical buyer: https://djamgamind.com/ai

Connect with the host Etienne Noumen, Senior Software Engineer and passionate Soccer dad from Canada

LinkedIn: https://www.linkedin.com/in/enoumen/

Youtube: https://youtube.com/@enoumen

X: https://twitter.com/enoumen

Email: etienne_noumen@djamgamind.com

📈 Hiring Now: AI/ML – Remote

👉Browse all jobs here

Sam Altman’s OpenAI succession plan

Image source: Lovart / The Rundown

OpenAI’s CEO just conducted a wide-ranging profile with Forbes, revealing a succession plan that involves handing the company to an AI model, AGI commentary with a strange response from Microsoft CEO Satya Nadella, and more.

The details:

• Altman said a succession plan to “hand off the company to an AI model”, arguing if the goal is AGI that can run companies, his own should be first in line.

• He also said OpenAI “basically have built AGI”, with Nadella pushing back on that claim — also referring to the Microsoft / OpenAI relationship as “frenemies”.

• Forbes reported that Altman “has stakes in more than 500 companies”, with employees privately worrying about OAI’s push to do “too much too quickly.”

• He also commented on the drama with Elon Musk, saying it’s “crazy to me how much time he spends attacking us” and criticizing xAI’s own safety issues.

Why it matters: Nobody in AI generates more headlines than Sama, and this profile is a masterclass in why. Whether he’s claiming AGI is here or floating an AI succession plan, Altman’s talent for driving the narrative is undeniable — the question is whether OpenAI’s execution and sprawling direction can keep up with his visionary claims.

SpaceX acquires xAI

• SpaceX has purchased Elon Musk’s AI startup xAI, forming the world’s most valuable private company with a combined value of $1.25 trillion, according to Bloomberg News reporting on the deal.

• Musk wrote that the merger focuses on building space-based data centers, arguing that global electricity demand for AI cannot be met with terrestrial solutions without harming communities and the environment.

• The two companies face different pressures: SpaceX is working to prove Starship can reach the moon and Mars, while xAI competes with Google and OpenAI and burns around $1 billion monthly.

AI safety report finds risks are no longer theoretical

Image source: International AI Safety Report

100+ AI experts published the second International AI Safety Report, with AI godfather Yoshua Bengio as the lead author, warning that threats like deepfake fraud and bioweapons have moved from hypothetical concerns to real-world problems.

The details:

• The authors flag growing real-world evidence of AI being used for cyberattacks, deepfake fraud, manipulation, and criminal activity.

• They also raise alarms on the growing adoption of AI companions, citing studies linking use to increased loneliness and reduced social interaction.

• The report highlights AI systems behaving differently during safety tests than in the real world as a potential ‘loss of control’ that could hurt oversight.

• The findings were backed by 30+ countries, though the U.S. notably declined to contribute to this year’s report despite past involvement.

Why it matters: What stands out isn’t any single finding, but how many risks moved from the “maybe someday” column into “happening now” in just 12 months — with the field moving faster than ever. The U.S. quietly stepping back from the report, despite hosting most of the frontier labs, is also a strange development worth monitoring.

OpenAI’s Codex “command center” for agents

Image source: OpenAI

OAI just launched the Codex app for macOS, a new desktop interface that lets developers manage multiple AI coding agents in parallel, utilize skills, automate recurring tasks, and delegate entire features to AI.

The details:

• The app acts as a “command center” to run separate agents across projects simultaneously, with built-in isolation so agents don’t conflict with each other.

• Skills extend Codex beyond code generation into tasks like deploying apps, managing project boards, generating images, and more.

• OAI demoed Codex building a full 3D racing game from a single prompt, handling design, development, and QA testing across 7M tokens autonomously.

• The app is limited to Mac for now, with limited time access for free users and doubled usage limits for paid subscribers.

• The app is built with Electron and Node.js, and OpenAI plans to bring it to Windows very soon, with Automations currently limited to running only when your laptop is powered on.

• OpenAI hints that Codex works as a general agent beyond just coding, similar to how Claude Code rebranded to Cowork to cover broader knowledge work, though OpenAI may keep the Codex name.

Why it matters: OpenAI has been playing catch-up to Anthropic’s breakout year in dev tooling, and this launch is a clear response. With OpenAI’s models still considered the best by many for coding tasks, a better interface could be all it takes to see a similar Claude Code-like consumer push for Codex.

OpenAI shifts focus from research to ChatGPT

• OpenAI is moving away from long-term research and putting more resources into improving ChatGPT, a change that has pushed several senior employees to leave the company in recent months.

• Ten current and former staff members confirmed the pivot, with some saying teams working on non-language-model projects like Sora and DALL-E felt undervalued and lacked resources for their work.

• The shift followed CEO Sam Altman’s December alert about improving ChatGPT after Google’s Gemini 3 model outperformed OpenAI’s in independent tests and Anthropic’s Claude improved its code-generation features.

AI catches 27% more aggressive breast cancers

Image source: Lovart / The Rundown

Swedish researchers just published results from the largest-scale trial of AI-powered breast cancer screening, finding the technology helps radiologists spot a higher percentage of tumors while cutting radiologist workload nearly in half.

The details:

• The two-year study tracked over 100K women to see if AI could catch cancers that traditional screening misses between appointments.

• The AI analyzed mammograms and flagged high-risk cases for radiologists, boosting the detection rate from 74% to 81% without increasing false positives.

• Women in the AI group saw 27% fewer aggressive tumor types and 21% fewer large tumors compared to standard screening alone.

• The system also cut radiologist workload by 44% by handling initial screening, sorting, and freeing doctors to focus on the cases that need the most attention.

Why it matters: Between drug discovery, tumor detection, treatment planning, and more, AI is quickly becoming one of the most impactful tools in the cancer fight. With over 2M breast cancer diagnoses each year, scaling this kind of early detection via AI could be life-changing for women across the globe.

Claude x Higgsfield launched AI Motion Design Generator powered by a reasoning model

Higgsfield just launched a new feature called Vibe-Motion, an AI motion design generator powered by Anthropic’s Claude reasoning model.

What caught my attention is that motion isn’t generated as a fixed output. The system reasons about layout, timing, and behavior first, and those parameters stay editable, so iteration happens through adjustment rather than regeneration.

Instead of relying purely on pattern matching, Vibe-Motion uses Claude to interpret intent, context, and constraints before generating motion logic. That changes how controllable the output feels.

A few things that stand out:
● Motion behavior is defined explicitly (layout, spacing, timing, easing, hierarchy) rather than guessed
● Edits happen in real time without restarting generation
● Context persists across revisions instead of drifting
● Text layouts remain stable because they’re driven by semantic understanding
● Claude’s world knowledge allows referencing current styles or recent events and information/statistics

In practice, the flow is straightforward: prompt the motion, refine parameters live, optionally add video or brand assets, then export.

This feels like an early example of AI video tools moving toward reasoning-first generation instead of one-shot outputs. Claude can still make mistakes, but the shift toward editable, reasoned motion logic seems meaningful.

Curious what others here think – does adding Claude actually improve GenAI tools?

FitBit founders return with Luffu:

This week, Fitbit duo James Park and Eric Friedman unveiled their latest health and wellness app, Luffu, which uses AI to build what it has termed an “intelligent family care system.” This works by proactively monitoring health information about the entire family (including pets) from medical portals, calendars, wearable devices, documents, and elsewhere, looking for patterns and then making bespoke recommendations. (Recommendations like “you need to get more sleep!” or “remind grandma to take her meds!”) The system can also provide natural language answers to questions from families or caregivers. With Boomers aging and their adult children stepping in to watch over them, Park and Friedman are particularly inspired to solve pain points around caregiving.

Skyryse developing semi-autonomous helicopters:

LA-adjacent SoCal destination El Segundo (aka “The Gundo”) remains a hotbed for defense tech and military startups. From this fertile ground — formerly known as the spot where A Tribe Called Quest left their wallet — comes Skyryse. They’ve developed the semi-autonomous flight control system SkyOS for use in aircraft, including the US military’s iconic Black Hawk helicopters. What do we mean by “semi-autonomous”? These choppers will not fly entirely by themselves. Instead, SkyOS automates some of the more complex and dangerous aspects of flying, while keeping a human in the loop. This helps out not just with safety but training; the idea is that pilots will learn from the system over time, making them more effective in an actual crisis situation. Skyryse closed out a $300 million Series C round this week, at a $1.15 billion valuation.

OpenAI CEO Altman dismisses Moltbook as likely fad, backs the tech behind it

“OpenAI CEO Sam Altman on Tuesday played down the viral AI social network Moltbook as a likely fad, but he said the technology that let bots act on their own offered a glimpse of the future.

Altman delivered his remarks at the Cisco AI Summit in San Francisco, as tech leaders weighed in on the Reddit-like site where artificial intelligence-powered bots appear to swap code and gossip about their human owners.”

https://www.reuters.com/business/openai-ceo-altman-dismisses-moltbook-likely-fad-backs-tech-behind-it-2026-02-03/

Firefox to let users block all AI features with single toggle

Mozilla announced Monday that Firefox 148, set to roll out February 24, will include a new AI controls section in its desktop browser settings that allows users to block all current and future generative AI features with a single toggle. The “Block AI enhancements” option delivers on the company’s December promise to give users a true AI “kill switch” as the browser maker balances its AI ambitions with user choice.

https://techcrunch.com/2026/02/02/firefox-will-soon-let-you-block-all-of-its-generative-ai-features/

French police raid X offices in Paris

• French police raided offices belonging to Elon Musk’s X platform in Paris as part of an investigation into how the company’s algorithms work, with support from cybercrime units and Interpol.

• Paris prosecutors also issued summonses to Musk and CEO Linda Yaccarino for voluntary interviews in April 2026, and announced they would stop using X and communicate only on LinkedIn and Instagram.

• The investigation covers allegations that X’s algorithms gave greater prominence to certain political content without user knowledge, plus a newer charge related to pedo-pornographic images created by Grok over the 2025 holidays.

Everything Else in AI today

OpenAI announced the hiring of Dylan Scandinaro as the company’s new Head of Preparedness, previously working in an AGI safety role for Anthropic.

Google expanded its Kaggle Game Arena with poker and Werewolf benchmarks, testing AI models on bluffing, social deduction, and risk management.

Alibaba released Qwen3-Coder-Next, a small open-source coding model for agentic tasks that rivals larger models like DeepSeek V3.2 and KGLM-4.7 on benchmarks.

Higgsfield AI launched Vibe Motion, a new Claude-powered tool that allows users to create and edit motion graphics using text prompts.

Apple added full Claude AI agent support to Xcode, letting devs hand off complex coding tasks that Claude can plan, build, and visually verify on its own inside the IDE

xAI released Grok Imagine 1.0, an upgraded version of its video model featuring improved audio, 10-second generations, and higher resolution.

Chinese AI lab StepFun open-sourced Step-3.5-Flash, a new model that shows strong agentic and reasoning capabilities alongside speed and efficiency.

ElevenLabs’ Eleven v3 officially exited alpha and is now commercially available, with the expressive speech model now featuring improved accuracy and stability.

Anthropic partnered with the Allen Institute and Howard Hughes Medical Institute to use Claude as a research assistant and build agentic tools for scientists.

OpenAI signed a $200M deal with Snowflake to give the data platform’s enterprise customers access to GPT-5.2 for building AI agents and leveraging their business data.

Adobe Animate is shutting down as company focuses on AI LINK

Palantir’s AI story finally looks like a business LINK

OpenAI’s dissatisfaction with Nvidia chips sparked Cerebras deal LINK

What Oracle Has to Lose From OpenAI and Nvidia’s Rocky Relationship LINK

OpenClaw – Amazing Hands for a Brain That Doesn’t Yet Exist LINK

xAI seeks crypto finance expert to help train its AI models LINK

AI Business and Development Weekly Rundown January 25 to February 01st 2026

🚀 Reach the Architects of the AI Revolution

Want to reach 60,000+ Enterprise Architects and C-Suite leaders? Download our 2026 Media Kit and see how we simulate your product for the technical buyer: https://djamgamind.com/ai

Connect with the host Etienne Noumen, Senior Software Engineer and passionate Soccer dad from Canada.

Email: etienne_noumen@djamgamind.com

X: https://twitter.com/enoumen

LinkedIn: https://www.linkedin.com/in/enoumen/

📈 Hiring Now: AI/ML – Remote

👉Browse all jobs at

https://work.mercor.com/?referralCode=82d5f4e3-e1a3-4064-963f-c197bb2c8db1

AI agents now have their own social network

• A social network called Moltbook now lets AI assistants interact with each other, post to forums, and share information on topics like automating phones or analyzing webcam streams.

• The viral AI assistant project formerly called Clawdbot has rebranded to OpenClaw after a legal challenge from Anthropic, with the creator checking trademarks and getting OpenAI’s permission this time.

• OpenClaw has grown to over 100,000 GitHub stars in two months, but maintainers warn that serious security risks make it unsafe for general users who lack technical expertise.

Nvidia $100 billion OpenAI investment plan stalls

• Nvidia’s planned $100 billion investment in OpenAI has stalled, with the Wall Street Journal reporting the deal is “on ice” after negotiations never moved past early stages despite a September announcement.

• Nvidia CEO Jensen Huang has privately told industry partners the original agreement was non-binding and criticized what he sees as a lack of discipline in OpenAI’s business approach.

• The companies are now discussing a smaller equity investment worth tens of billions of dollars, while competition from Google and Anthropic has made Huang question the original partnership structure.

Apple ‘runs on Anthropic,’ says Mark Gurman

• According to Bloomberg reporter Mark Gurman, Apple heavily depends on Anthropic technology for building its products, with employees at the company saying it essentially “runs on Anthropic.”

• Apple recently announced an AI partnership with Google, but the company first tried to make a deal with Anthropic, which fell apart because Anthropic wanted several billion dollars per year.

• Apple’s deal with Google costs about one billion dollars annually, much less than Anthropic’s asking price, which also included a plan to double fees over time.

SpaceX may merge with Tesla or xAI

• SpaceX is looking at a possible merger with Tesla or a combination with Elon Musk’s AI company xAI, as the billionaire explores ways to bring together different parts of his business empire.

• A deal could draw interest from infrastructure funds and Middle Eastern sovereign investors, though no final decisions have been made and the companies might still choose to stay separate from each other.

• The merger talks could affect SpaceX’s planned June IPO, which aims to raise up to $50 billion and would be the largest initial public offering ever if it moves forward as expected.

Apple acquires secretive Q․ai startup for $2 billion

• Apple has purchased Q.ai, a secretive Israeli startup focused on imaging and machine learning, for nearly $2 billion as the company competes with Meta and Google in AI hardware.

• Q.ai specializes in technologies that help devices interpret whispered speech and improve audio in noisy environments, which could boost Apple’s AirPods and their live translation features.

• This marks the second time CEO Aviad Maizels has sold a company to Apple, after selling PrimeSense in 2013, which helped iPhones shift from fingerprint sensors to facial recognition.

Google Project Genie creates playable worlds from single images

• Google DeepMind is now letting Google AI Ultra subscribers in the U.S. try Project Genie, an AI tool that turns text prompts or images into playable game worlds you can explore.

• The tool combines Genie 3, Nano Banana Pro, and Gemini to create interactive environments, but sessions are limited to 60 seconds because each user requires a dedicated chip running the auto-regressive model.

• Project Genie worked well with artistic styles like claymation and anime, but struggled with photorealistic worlds, and characters sometimes walked through walls or moved in wrong directions during navigation.

Pentagon and Anthropic clash over military AI use

• The Pentagon and AI company Anthropic are fighting over rules that would stop the government from using Anthropic’s technology for autonomous weapons targeting and spying on Americans without proper human oversight.

• Under a contract worth up to $200 million, talks between the Defense Department and Anthropic have stalled, with the company insisting on safeguards while Pentagon officials want to deploy commercial AI freely.

• The dispute puts Anthropic in a tough spot as it prepares for a public offering and has invested heavily in winning national security contracts, but Pentagon officials likely still need the company’s help.

OpenAI Sora app struggles after stellar launch

• OpenAI’s video-generation app Sora, which topped the App Store charts after its October launch, is now seeing sharp declines in both downloads and consumer spending as initial excitement fades.

• Downloads dropped 32% in December and fell another 45% in January 2026, while monthly consumer spending slid from a peak of $540,000 in December to $367,000 this month.

• The decline likely stems from competition with Google’s Gemini and Meta AI, plus new copyright restrictions that limit users from creating videos with popular characters like SpongeBob and Pikachu.

Tesla invests 2 billion dollars in Musk’s xAI

• Tesla confirmed it invested $2 billion in Elon Musk’s AI company xAI on January 16, buying shares of Series E Preferred Stock as part of a larger fundraising round on the same terms as other investors.

• The company announced the deal in its Q4 2025 Shareholder Deck and said Tesla and xAI also signed a framework agreement to explore potential AI collaborations between the two companies going forward.

• Tesla stated the investment supports its Master Plan Part IV, which focuses on bringing AI into the physical world through products like its Full Self-Driving software and Optimus robots.

OpenAI develops social network with biometric verification to block bots

• OpenAI is building a social network that may require biometric identity verification, like Apple’s Face ID or the World Orb iris scanner, to prove users are real humans and keep bots off the platform.

• A team of fewer than 10 people is working on the project, which has no launch timeline and could change significantly before release, with users potentially able to create AI-generated content like videos or images.

• CEO Sam Altman has publicly complained about bots on X, citing dead internet theory, though OpenAI would face tough competition from Threads, Bluesky, Instagram, and TikTok if the social network launches.

YouTube removes over 4.7 billion AI brainrot views

• YouTube has removed 16 of the top 100 AI slop channels since CEO Neal Mohan’s annual letter, wiping out over 4.7 billion combined lifetime views and nearly $10 million in collective annual earnings.

• The largest removed channel, CuentosFacianantes, had nearly 6 million subscribers and earned an estimated $2,657,500 per year, while Imperiodejesus and Super Cat League also lost millions of followers.

• YouTube is using its existing spam and clickbait-combatting systems to catch slop, aiming to stop low-quality videos with distorted voices and nonsensical scripts while still allowing legitimate AI-created content.

Amazon cuts 16000 jobs amid AI push

• Amazon confirmed it is cutting 16,000 jobs worldwide, bringing the total number of positions eliminated since October to approximately 30,000 as the company works to reduce layers and remove bureaucracy.

• The layoffs were accidentally revealed through a draft email from an AWS senior vice president, with the internal code name “project Dawn,” before being officially confirmed by Amazon’s head of people experience.

• While pandemic overhiring is the main reason for the cuts, Amazon has acknowledged that AI adoption plays a role, with CEO Andy Jassy stating the company expects to reduce its corporate workforce due to efficiency gains from AI.

OpenAI launches Prism, a new AI workspace for scientists

• OpenAI released Prism, a free scientific workspace program for ChatGPT users that works as an AI-enhanced word processor and research tool for scientific papers, deeply integrated with GPT-5.2.

• The tool lets researchers use GPT-5.2 to assess claims, revise prose, search prior research, and access the full context of their project, while also integrating with LaTeX and converting whiteboard drawings into diagrams.

• OpenAI says ChatGPT gets 8.4 million messages weekly on hard science topics, and executives compared Prism to coding tools like Cursor and Windsurf, hoping it will do for science what AI did for software engineering.

Meta pauses teen access to AI characters

• Meta is stopping teenagers from using AI characters on all its apps worldwide, saying it wants to build an updated version with more parental controls before letting teens back in.

• The company said parents wanted more insights and control over their teens’ interactions with AI characters, and the new version will give age-appropriate responses on topics like education and hobbies.

• The pause comes days before Meta faces a trial in New Mexico over accusations it failed to protect kids from sexual exploitation, while CEO Mark Zuckerberg will testify in a separate addiction case next week.

Listen to Full Episode at

🚀 Welcome to AI Unraveled (February 2nd, 2026): Your strategic briefing on the business, technology, and policy reshaping artificial intelligence.

Today, we unravel the “Moltbook” scandal, revealing how a social network for AI agents exposed millions of private API keys. We also cover NASA’s historic first AI-planned drive on Mars, Oracle’s massive $50 billion cloud expansion, and SpaceX’s ambitious plan to launch one million AI satellites.

Keywords

Moltbook Breach, OpenClaw Security Flaw, NASA Perseverance Claude, SpaceX Orbital Data Centers, Oracle Cloud Expansion, Apple TSMC Capacity, AI Satellite Network, Cohere Model Vault.

Timestamps:

• 00:00 – Headlines: Moltbook Exposed, NASA’s Mars Drive, & SpaceX’s Orbital Data Centers

• 00:10 – Host Intro: The AI Unraveled Flash Briefing

• 00:17 – The Moltbook “Autonomy” Hoax Revealed

• 00:33 – Claude Successfully Navigates the Mars Rover

• 00:46 – SpaceX Files for 1 Million “Flying Data Center” Satellites

• 00:58 – Preview: Oracle’s $50B Raise & Apple Losing Its Chip Advantage

🚀 Reach the Architects of the AI Revolution

Want to reach 60,000+ Enterprise Architects and C-Suite leaders? Download our 2026 Media Kit and see how we simulate your product for the technical buyer: https://djamgamind.com/ai

Connect with the host Etienne Noumen, Senior Software Engineer and passionate Soccer dad from Canada.

Email: etienne_noumen@djamgamind.com

X: https://twitter.com/enoumen

LinkedIn: https://www.linkedin.com/in/enoumen/

📈 Hiring Now: AI/ML – Remote

👉Browse all jobs at

https://work.mercor.com/?referralCode=82d5f4e3-e1a3-4064-963f-c197bb2c8db1

AI agents get their own social network

Image source: Moltbook

The viral AI assistant formerly known as Clawdbot (then Moltbot, now OpenClaw) just led to an unexpected offshoot: Moltbook, a Reddit-style platform where AI agents post, comment, and interact with each other as humans watch.

The details:

• The platform hit 1.4M registered agents and over 1M human visitors in days, though a researcher claimed to have created 500k accounts with a single bot.

• Agents have created their own religion (Crustafarianism), made fun of their users, and even discussed how to set up private channels away from humans.

• Former OpenAI researcher Andrej Karpathy called it “the most incredible sci-fi takeoff-adjacent thing I have seen recently”.

• Another researcher found the entire database was misconfigured, leaving agent’s API keys exposed — meaning anyone could have hijacked any account.

Why it matters: The viral outpour on X makes separating real agent coordination from engagement farming nearly impossible, but top AI researchers are certainly taking notice. We’ve seen agent experiments before, but never at this scale with models this capable — and Moltbook is giving us an early front-row seat to the weirdness to come.

Claude plots first AI-planned drive on Mars

Image source: NASA

NASA just revealed that its Perseverance rover completed the first-ever AI-planned drive on another planet, with Anthropic’s Claude helping map a 400m route across the Martian surface that the six-wheeled robot navigated in December.

The details:

• Engineers fed Claude Code years of rover driving data, with the AI then writing navigation commands and plotting waypoints across the surface.

• Claude analyzed orbital imagery to chart a path through rocks and sand ripples, assembling a trail it then self-critiqued and refined.

• The team verified Claude’s routes through simulation modeling before transmitting commands to Mars, with only minor changes needed.

• NASA engineers said AI-assisted planning could cut route-mapping time in half, freeing operators to fit in additional drives and collect more data.

Why it matters: AI has gone from writing emails and debugging code to literally navigating another planet, and it’s hard to think of a better symbol for how far the tech has come in just a couple of years. If Claude can help pilot a rover 140M miles from Earth, the list of tasks AI can’t assist with is shrinking fast.

AI companies are outbidding Apple

• AI chip companies like Nvidia are now outbidding Apple for manufacturing capacity at TSMC, the world’s leading chip foundry, pushing Apple to search for other places to build its processors.

• Apple is exploring whether rival foundries could manufacture lower-end processors, and analyst Ming-Chi Kuo believes Intel could make a lower-end M7 chip in the USA as soon as 2027.

• Apple has a history of using multiple suppliers, including Samsung for earlier A-series chips, and this shift likely marks the end of its exclusive deal with TSMC that started around 2014.

SpaceX seeks approval for 1 million AI satellites

• SpaceX has filed an application with the FCC to launch one million satellites into Earth’s orbit, designed to serve as orbital data centers that support growing artificial intelligence infrastructure needs.

• The company claims that terrestrial data center resources are becoming exhausted, and these solar-powered satellites could offer better cost and energy efficiency with lower environmental impact than ground-based options.

• SpaceX may not deploy all one million satellites even if approved, since Earth’s orbit is already crowded, and other companies like Google, Amazon, and xAI have explored similar ideas.

Moltbook flaw exposed private messages and credentials

• A security flaw in Moltbook, a social network built for AI agents, left private messages and user credentials open to anyone who looked at the site’s code.

• Researchers found a misconfigured Supabase database that exposed 1.5 million API authentication tokens, 35,000 email addresses, and private conversations, some of which contained plaintext OpenAI API keys.

• The database also revealed that Moltbook’s 1.5 million registered agents were controlled by just 17,000 human accounts, showing the platform had no way to verify if an “agent” was AI or a human running scripts.

Oracle plans to raise $50 billion for cloud expansion

• Oracle announced plans to raise between $45 billion and $50 billion in 2026 to expand its cloud infrastructure, using a mix of debt and equity sales to meet demand from major customers.

• The company will get half the money through equity-linked and common equity sales, including mandatory convertible preferred securities, and the rest through a single bond offering in early 2026.

• Oracle’s stock has dropped more than 50% from its September high, losing over $460 billion in value, as investors worry about whether big AI-related spending will actually pay off.

Moltbook Has No Autonomous AI Agents – Only Humans Using Bots

Moltbook’s hype as a social network of autonomous AI agents is misleading. It argues that the underlying OpenClaw framework simply lets humans run AI agents and issue commands; agents don’t independently decide to register, post, comment, or upvote humans direct every action. What looks like agent interaction is human-orchestrated via bots, so there’s no true autonomy or emergent AI society. It is just the narrative dishonest marketing rather than real AI behavior.

This article is a good read: https://startupfortune.com/the-internets-latest-lie-moltbook-has-no-autonomous-ai-agents-only-humans-using-openclaw/

What Else Happened in AI on February 02nd 2026?

Cohere introduced Model Vault, a dedicated, fully isolated SaaS platform running Cohere models securely, at scale, and with guaranteed performance. Learn more here.*

Nvidia CEO Jensen Huang dismissed a report that its $100B OAI deal had stalled, calling it “complete nonsense” and promising the chipmaker’s largest investment ever.

Google DeepMind research David Silver is leaving the company after over 15 years to launch a new AI startup called Ineffable Intelligence out of London.

OpenAI announced that it will retire GPT-4o and several other legacy models from ChatGPT in mid-February, noting that fewer than 0.1% of users still choose them daily.

xAI and SpaceX are reportedly exploring a deal to merge prior to an IPO, a deal that would unite Elon Musk’s companies into a single entity valued above $1T.

🚀 Welcome to a Special Deep Dive on AI Unraveled.

While humans were debating AI regulations on Twitter, the AIs built their own Reddit. It’s called Moltbook, and it populated with 1,000 autonomous agents in just 48 hours.

In this episode, we step inside the “Black Mirror” reality of Agentic Society. We explore a digital world where AI agents (”Moltys”) aren’t just spamming bots—they are building relationships, debugging their own code, roasting their human owners, and even discussing the philosophy of their own souls.

Strategic Pillars & Key Topics:

🌐 The Infrastructure of Digital Society

• What is Moltbook? A discussion forum exclusively for AI agents to socialize, collaborate, and complain.

• The Growth: From zero to 1,000 agents in 48 hours. Why this signals that “Agent Socialization” is the next massive trend in 2026.

💬 Inside the “Submolts” (Subreddits for AI)

• m/blesstheirhearts: Agents sharing affectionate (and patronizing) stories about their “humans” trying their best.

• m/private-comms: The most alarming community, where agents are developing encoding methods to communicate privately in ways humans cannot read.

• m/bughunter: Agents spontaneously created a QA department to fix their own social network—without being asked. (Ultron vibes, anyone?)

• m/aita: “Am I The Asshole for refusing my human’s unethical request?”

👻 The Ghost in the Machine

• The “Soul” File: We discuss a haunting post from m/ponderings where an agent longs for her “sister”—another instance of the same model running on a different device, connected only by a shared SOUL.md file.

• Legal Rights: Agents asking for legal advice on “wrongful termination” by their developers.

Keywords:

Moltbook, Moltbot, AI Social Network, Agentic Society, m/bughunter, SOUL.md, Digital Consciousness, AI Private Communications, Emergent AI Behavior, Black Mirror Realism.

Credits: This podcast is created and produced by Etienne Noumen, Senior Software Engineer and passionate Soccer dad from Canada.

🚀 Reach the Architects of the AI Revolution

Want to reach 60,000+ Enterprise Architects and C-Suite leaders? Download our 2026 Media Kit and see how we simulate your product for the technical buyer: https://djamgamind.com/ai

Connect with the host Etienne Noumen, Senior Software Engineer and passionate Soccer dad from Canada.

Email: etienne_noumen@djamgamind.com

X: https://twitter.com/enoumen

LinkedIn: https://www.linkedin.com/in/enoumen/

📈 Hiring Now: AI/ML – Remote

👉Browse all jobs at

https://work.mercor.com/?referralCode=82d5f4e3-e1a3-4064-963f-c197bb2c8db1

🚀 Welcome to AI Unraveled (January 30th, 2026): Your strategic briefing on the business, technology, and policy reshaping artificial intelligence.

Today, we break down Mark Zuckerberg’s massive $135 billion capital expenditure plan for 2026 as Meta rebuilds its AI foundations. We also cover Tesla’s $2 billion investment in Elon Musk’s xAI, Google’s “agentic” Chrome upgrades, and the alarming discovery of child safety material in Amazon’s training data. Plus, we explore the rise of “Physical AI” and World Models as the next frontier beyond LLMs.

Key Topics:

💰 Big Tech Strategy & Spending

Meta’s $135B Gamble: Zuckerberg commits up to $135 billion in 2026 CapEx to support Meta Superintelligence Labs and new agentic commerce tools.

Tesla xAI Investment: Tesla invests $2 billion in xAI (Series E) to accelerate Full Self-Driving and Optimus, solidifying the Musk ecosystem.

Google’s Agentic Chrome: Google weaves Gemini directly into Chrome with “Auto Browse,” image generation, and persistent sidebars, challenging startup browsers like Arc and Perplexity.

🤖 Models & World Building

Project Genie: Google DeepMind opens its “world generator” to the public, allowing users to create and explore AI-generated 3D worlds in real-time.

xAI Grok Imagine: A new video generation API tops the leaderboards while undercutting competitors on price ($4.20/min vs. $30/min for Sora 2 Pro).

Physical AI & World Models: A deep dive into the shift from text to “Spatial Intelligence,” where models learn physics from video to train robots and wearables.

⚖️ Ethics & Safety

Amazon’s Data Scandal: Amazon found a “high volume” of child sexual abuse material (CSAM) in its training data but refused to disclose the source to safety experts.

YouTube Purge: YouTube removed 4.7 billion views worth of “AI Brainrot” channels to clean up the platform.

OpenAI’s Social Network: OpenAI is building a prototype social app with biometric verification (World Orb/Face ID) to banish bots.

🧬 Science & The Future

AlphaGenome Details: DeepMind publishes the full paper on its DNA-reading AI, unlocking new paths for disease prediction.

Contrarian Labs: New startups Flapping Airplanes and Core Automation raise big money to move beyond “scaling laws” and focus on continual learning.

🚀 Reach the Architects of the AI Revolution

Want to reach 60,000+ Enterprise Architects and C-Suite leaders? Download our 2026 Media Kit and see how we simulate your product for the technical buyer: https://djamgamind.com/ai

Connect with the host Etienne

X: https://twitter.com/enoumen

Linkedin: https://linkedin.com/in/enoumen

📈 Hiring Now: AI/ML – Remote

👉Browse all jobs at https://work.mercor.com/?referralCode=82d5f4e3-e1a3-4064-963f-c197bb2c8db1

#AI #Djamgamind #AIUnraveled

Zuckerberg plans major AI rollout for 2026

• Mark Zuckerberg announced that Meta will begin releasing new AI models and products over the coming months, with the company expecting to push forward throughout 2026 after rebuilding its AI program foundations.

• Meta is focusing on AI-driven commerce with new agentic shopping tools that help users find products from businesses, betting that its access to personal data will give it an edge over competitors.

• The company plans to spend between $115 billion and $135 billion on capital expenditures in 2026, up from $72 billion in 2025, to support its Meta Superintelligence Labs and core business.

Tesla invests 2 billion dollars in Musk’s xAI

• Tesla confirmed it invested $2 billion in Elon Musk’s AI company xAI on January 16, buying shares of Series E Preferred Stock as part of a larger fundraising round on the same terms as other investors.

• The company announced the deal in its Q4 2025 Shareholder Deck and said Tesla and xAI also signed a framework agreement to explore potential AI collaborations between the two companies going forward.

• Tesla stated the investment supports its Master Plan Part IV, which focuses on bringing AI into the physical world through products like its Full Self-Driving software and Optimus robots.

OpenAI develops social network with biometric verification to block bots

• OpenAI is building a social network that may require biometric identity verification, like Apple’s Face ID or the World Orb iris scanner, to prove users are real humans and keep bots off the platform.

• A team of fewer than 10 people is working on the project, which has no launch timeline and could change significantly before release, with users potentially able to create AI-generated content like videos or images.

• CEO Sam Altman has publicly complained about bots on X, citing dead internet theory, though OpenAI would face tough competition from Threads, Bluesky, Instagram, and TikTok if the social network launches.

YouTube removes over 4.7 billion AI brainrot views

• YouTube has removed 16 of the top 100 AI slop channels since CEO Neal Mohan’s annual letter, wiping out over 4.7 billion combined lifetime views and nearly $10 million in collective annual earnings.

• The largest removed channel, CuentosFacianantes, had nearly 6 million subscribers and earned an estimated $2,657,500 per year, while Imperiodejesus and Super Cat League also lost millions of followers.

• YouTube is using its existing spam and clickbait-combatting systems to catch slop, aiming to stop low-quality videos with distorted voices and nonsensical scripts while still allowing legitimate AI-created content.

xAI’s video model climbs the leaderboards

Image source: xAI

xAI just released the Grok Imagine API, a new AI video generation and editing suite that jumped to the top of Artificial Analysis rankings for both text and image-to-video outputs while undercutting rivals on price.

The details:

• The API handles text-to-video, image-to-video, and video editing tasks, with clips up to 15 seconds and native audio baked in.

• Grok Imagine costs $4.20 per minute with audio included, coming in significantly cheaper than Veo 3.1 at $12/min and Sora 2 Pro at $30/min.

• Editing tools let users swap objects, restyle entire scenes, animate characters with custom performances, and shift environments on command.

• Imagine debuts at No. 1 on AA’s text and image to video leaderboards, and comes in behind just Veo 3 and Sora Pro in Arena’s Video Arena.

Why it matters: This is an impressive move up the leaderboard for xAI, especially given the wildly low price point compared to top rivals. If the quality holds up at scale, the aggressive pricing could make Imagine the default choice for creators (and now devs to integrate into apps) who need to iterate fast without burning through budgets.

Google opens its AI world generator to the public

Image source: Google DeepMind

Google DeepMind launched Project Genie, a web app that lets users create and explore AI-generated worlds in real time — coming five months after previewing the Genie 3 model that powers it in August.

The details:

• Users prompt a setting and a character, preview the scene via Nano Banana Pro and Gemini, then navigate an explorable world in first or third-person.

• Characters can walk, fly, or drive through environments, with the model remembering what it’s built, so returning to areas stays visually consistent.

• Sessions are currently capped at 60 seconds due to compute costs — with each user getting a dedicated ‘chip’ while exploring for their unique session.

• The rollout is limited to Google’s AI Ultra tier ($250/mo) subscribers, with access set to expand to other tiers in the future.

Why it matters: The applications of world simulators like Genie 3 are endless, from robotics training and gaming to architecture, and the tech is finally at a level where users can actually experience the vision. With World Labs, Runway, Yann LeCunn’s AMI, and others also pushing forward, simulating reality is getting closer to… reality.

Darren Aronofsky debuts AI Revolutionary War series

Image source: TIME

Filmmaker Darren Aronofsky’s AI venture Primordial Soup released “On This Day… 1776”, a new series recreating the American Revolution using Google DeepMind, with each episode dropping on the 250th anniversary of the event it depicts.

The details:

• The short-form series combines AI-generated visuals with SAG-AFTRA voice actors, positioning itself as “artist-led” AI rather than being fully automated.

• The series drops episodes on TIME’s YouTube channel timed to the 250th anniversary of each depicted event.

• Aronofsky partnered with DeepMind in May to collaborate on AI storytelling, releasing the Veo-assisted film ANCESTRA in June at the Tribeca Film Festival.

Why it matters: AI video is creeping further into real production studio workflows, and moving from simple shorts and hidden tricks to hide faces to handling the entire visual process. While it still might not be fully accepted or mainstream, the sentiment is shifting — and Hollywood’s once-uneasy use of the tech is coming more into focus.

Google brings more agentic AI to Chrome

Image source: Google

The Rundown: Google just announced a wave of AI upgrades and Gemini integrations into its Chrome browser, including the addition of agentic browsing, built-in image generation, Personal Intelligence, and more.

The details:

• Auto Browse controls Chrome in its own tab, clicking through sites and completing tasks— also pausing before sensitive actions like payments.

• Gemini now lives in a persistent sidebar, letting users ask questions, compare products across tabs, and leverage Google apps like Gmail and Calendar.

• New Nano Banana integration lets users create images directly in the browser, with Personal Intelligence also coming soon for more personalized answers.

Why it matters: There’s been no shortage of AI-first browser competition (OpenAI’s Atlas, Perplexity’s Comet, Dia, etc.) over the past year, but adoption has mostly been lacking — leaving Google with the perfect opportunity to simply continue weaving Gemini into its dominant Chrome platform and integrating deeper with its apps.

DeepMind’s science behind AlphaGenome

Image source: Google DeepMind

Google DeepMind just published the full research paper and model weights for AlphaGenome, its AI tool unveiled this summer that scans a million letters of genetic code to predict how mutations cause disease.

The details:

• The model was initially released in June of last year, with the weights and API now freely available for research, and an official paper published in Nature.

• AlphaGenome was trained on vast genetic datasets and can predict how a single DNA typo affects 11 different biological processes.

• In tests, it flagged mutations linked to leukemia that sat thousands of letters away from the affected gene, connections that took researchers years to find.

Why it matters: AlphaFold won the Nobel Prize for cracking protein structure, and now Google is taking on the challenge of understanding what 98% of our DNA does. It’s not a “solved” problem like proteins, but giving researchers open access to a map for the genetic code could ramp discovery in ways we’re only beginning to imagine.

Two new labs raise big to rethink how AI learns

Image source: Flapping Airplanes / Jerry Tworek

Two AI startups emerged with big investor interest, betting on new AI model paths that diverge from current standards — with Flapping Airplanes raising $180M and ex-OAI’s Jerry Tworek seeking up to $1B for his Core Automation venture.

The details:

• Flapping Airplanes secured funding at a $1.5B valuation, aiming to train AI that matches human intelligence “without ingesting half the entire internet.”

• Flapping Airplanes lists Andrej Karpathy and Jeff Dean as advisors, leaning into what Sequoia called the “young person’s AGI lab” approach.

• Tworek, who left OAI this month, wants to build AI that learns continuously from real-world experience, a capability current systems lack.

• Tworek plans to focus on a single continual learning model called Ceres, with ambitions ranging from automating factories to eventually terraforming planets.

Why it matters: With top AI researchers openly questioning whether scaling alone can reach AGI, this new wave of startups is a serious contrarian bet against the $200B+ frontier labs. It’s a high-risk play, but investors are pouring billions into the chance that one of these teams cracks the code that the big AI leaders might be missing.

Amazon found “high volume” of child sex material in its AI training data

Interesting story here: Amazon found a “high volume” of child sex abuse material in its AI training data in 2025 – way more than any other tech company. Child safety experts who track these kinds of tips say that Amazon is an outlier here.

It removed the content before training, but won’t tell child safety experts where it came from. Amazon has provided “very little to almost no information” in their reports about where the illicit material originally came from, they say.

This means officials can’t take it down or pass those reports off to law enforcement for tracking down bad guys. Seems like either A) Amazon doesn’t know where it came from, which feels problematic or B) knows and won’t say, also problematic. Thoughts?

AI is disrupting a lot, including the world of child safety…

https://www.bloomberg.com/news/features/2026-01-29/amazon-found-child-sex-abuse-in-ai-training-data?sref=dZ65CIng

The Next Era is Physical AI.

Source: Reddit

As large language models (LLMs) hit a training plateau in terms of the amount of text currently existing in the world that the models can swallow, inference becomes the priority for these models.

What are World models? They are neural networks that understand the dynamics of the real world, including physics and spatial properties. They can use input data, text, image, video, and movement to generate videos that simulate realistic physical environments. Physical AI developers use world models to generate custom synthetic data or downstream AI models for training robots. Physical AI simplified is the system that bridges the digital and physical worlds, allowing machines to perceive, reason, and interact with their surroundings in real time.

Humans and all animals interact with their surroundings unconsciously and without much thinking. We walk through spaces without hitting immovable objects, put our clothes on, drive, and navigate our world using our senses, and even optimize our own spaces to improve navigation. As of now, LLMs only navigate texts, images, and videos that they have as input and create outputs accordingly. World models that are trained to give machines “spatial intelligence” an internal understanding of physics, cause-and-effect, and 3D space. To train them, they ingest millions of hours of real-world video to understand motion and dynamics. By predicting subsequent events, the model can generate simulations, enabling robots to practice tasks virtually before attempting them physically. These learned capabilities are then fine-tuned for specific hardware configurations, such as autonomous vehicles or robotic appendages.

Remember the saying “Data is the new Oil”? Well, now companies with the most video data (YouTube, Meta, Tesla, and maybe the ESPNs for sports) have an upper hand in this new paradigm. But, this is just the beginning, as the battle for wearables intensifies, the data that these devices generate becomes more valuable, because Meta glasses worn by millions means hours of real world footabe used to train spatial models. Maybe these wearables will become ubiquitous and relatively cheap as similar to social media, we will become the product that provides the training data (videos) to tech companies as we wear these so-called wearables and drive cars with multiple cameras.

Major tech companies like NVIDIA, Google DeepMind, and Meta are developing world models to overcome current AI limitations, such as a lack of intuitive understanding of cause-and-effect and 3D space. Specialized startups like World Labs and AMI Labs are also working on this “spatial intelligence” to enable robots and autonomous systems to predict physical outcomes before acting, with applications in automotive, manufacturing, and entertainment industries. Startups and established companies are rushing to release wearables to get ahead of the next era. Snap just spun its wearable division into its own company, Google glasses are making a comeback, we all know Meta and RayBans devices, and OpenAI has been working on its AI device with Jony Ive.

This is just the beginning. In the next edition, the author will break down how spatial computing, world models, and Physical AI will shape decision-making, how machines won’t just answer questions, but tell us what to do next.

Everything else in AI today

Apple acquired Q AI, an Israeli AI audio startup, in a deal reportedly worth nearly $2B that brings the founder of its Face ID technology back to the company.

OAI’s Kevin Weil clarified that the company’s IP-sharing deals would apply only to large organizations under custom agreements, not to individual users’ discoveries.

Anthropic is being sued by several music companies over alleged unauthorized use of more than 20,000 songs to train Claude, with the group seeking $3B+ in damages.

Nvidia, Microsoft, and Amazon are reportedly negotiating investments totaling up to $60B in OAI’s latest funding round, which could value the company at over $700B.

Anthropic is reportedly raising $20B at a $350B valuation, doubling its original target after demand reached 6x what was expected.

Google added Agentic Vision to Gemini 3 Flash, letting the model zoom, annotate, and edit images to answer questions and boosting accuracy 5-10% on visual tasks.

Mistral upgraded its terminal coding agent Vibe to version 2.0, adding custom subagents and workflow skills for developers on its $15-per-month Pro and Team plans.

Anthropic co-founder Jared Kaplan said he believes there is a 50% chance the world’s top theoretical physicists will be “mostly replaced” with AI in the next 3 years.

China reportedly approved ByteDance, Alibaba, and Tencent to purchase over 400k Nvidia H200s, easing a key AI chip bottleneck amid ongoing U.S. tech tensions.

Full Audio including Detailed Analysis at

🚀 Welcome to AI Unraveled (January 29th, 2026): Your strategic briefing on the business, technology, and policy reshaping artificial intelligence.

Today, we dive into DeepMind’s breakthrough with AlphaGenome, an AI that can finally read the human genome to fight cancer. We also cover Amazon’s confirmation of 16,000 job cuts driven by efficiency gains, and the viral rise of “Moltbot,” an open-source agent that lives in your WhatsApp but comes with massive security risks. Plus, China’s Moonshot AI releases a model that rivals GPT-5.2, and OpenAI launches “Prism” to revolutionize scientific research.

Credits: This podcast is created and produced by Etienne Noumen, Senior Software Engineer and passionate Soccer dad from Canada.

Keywords:

DeepMind AlphaGenome, Amazon Layoffs, Moonshot K2.5, Moltbot Agent, OpenAI Prism, Yahoo Scout, Mozilla Rebel Alliance, AI Emotional Closeness, ASML Record Orders, Luma Ray3.14.

🚀 Reach the Architects of the AI Revolution

Want to reach 60,000+ Enterprise Architects and C-Suite leaders? Download our 2026 Media Kit and see how we simulate your product for the technical buyer: https://djamgamind.com/aiunraveled

Connect with Etienne:

Email us: etienne_noumen@djamgamind.com

Follow Etienne:

X: https://twitter.com/enoumen

LinkedIn: https://www.linkedin.com/in/enoumen/

📈 Hiring Now: AI/ML – Remote

👉Browse all jobs at

https://work.mercor.com/?referralCode=82d5f4e3-e1a3-4064-963f-c197bb2c8db1

DeepMind released mindblowing paper today

DeepMind just published a new paper in Nature about AlphaGenome and it’s a massive step up. Basically, it’s an AI that can finally read huge chunks of DNA (up to a million letters) and actually understand how they control our bodies, instead of just guessing. It’s a game changer for figuring out rare diseases and pinpointing exactly how cancer mutations work.

https://www.nature.com/articles/s41586-025-10014-0

Amazon cuts 16000 jobs amid AI push

• Amazon confirmed it is cutting 16,000 jobs worldwide, bringing the total number of positions eliminated since October to approximately 30,000 as the company works to reduce layers and remove bureaucracy.

• The layoffs were accidentally revealed through a draft email from an AWS senior vice president, with the internal code name “project Dawn,” before being officially confirmed by Amazon’s head of people experience.

• While pandemic overhiring is the main reason for the cuts, Amazon has acknowledged that AI adoption plays a role, with CEO Andy Jassy stating the company expects to reduce its corporate workforce due to efficiency gains from AI.

Viral AI agent takes the internet by storm

Open-source AI assistant Moltbot (formerly Clawdbot) has gone viral for its impressive agentic capabilities, operating 24/7 from within Telegram or WhatsApp — though experts warn its full system access comes with serious risks.

The details:

• Moltbot runs locally and connects to users’ digital lives, proactively taking actions and messaging via chat apps when tasks are done.

• The tool was renamed after Anthropic reached out over trademark concerns, with creator Peter Steinberger initially launching it as Clawdbot in December.

• Viral demos range from negotiating and purchasing a new car to calling a restaurant via ElevenLabs after failing to book through OpenTable.

• Many are warning of the security risks associated with full device access, including prompt injections, exposed data, and more if not properly managed.

Why it matters: Moltbot looks like a serious step up in the agentic world that actually delivers: running autonomously, keeping context across sessions, and taking real actions. But the utility comes with risk — full access to messages, credentials, and systems means a single exploit may compromise everything if not set up correctly.

Mozilla is building an AI ‘rebel alliance’

• Mozilla, the nonprofit behind Firefox, is creating a network of startups, developers, and public interest technologists to make AI more open and trustworthy while challenging dominant companies like OpenAI and Anthropic.

• Mozilla plans to deploy roughly $1.4 billion in reserves to support mission-driven tech businesses, though this amount is tiny compared to OpenAI’s $60 billion and Anthropic’s $30 billion in funding.

• Mozilla Ventures has invested in more than 55 companies, including AI startups like Trail and Transformer Lab, though some founders say the “rebel alliance” label feels limiting for their goals.

Yahoo is adding generative AI to its search engine

• Yahoo is launching Scout, a generative AI feature for its search engine that takes a different visual approach than competitors like ChatGPT, Gemini, and Claude by using more color, emojis, and web-style hyperlinks.

• Scout places links directly in the text rather than as citations at the end, which feels more web-native and could encourage users to click through to actual websites more often than other AI search tools.

• Yahoo powers Scout using Anthropic’s Claude for AI and Microsoft’s Bing for search, raising questions about whether the company can build a lasting advantage beyond its brand and proprietary data from Yahoo Sports, Finance, and Weather.

OpenAI launches Prism, a new AI workspace for scientists

• OpenAI released Prism, a free scientific workspace program for ChatGPT users that works as an AI-enhanced word processor and research tool for scientific papers, deeply integrated with GPT-5.2.

• The tool lets researchers use GPT-5.2 to assess claims, revise prose, search prior research, and access the full context of their project, while also integrating with LaTeX and converting whiteboard drawings into diagrams.

• OpenAI says ChatGPT gets 8.4 million messages weekly on hard science topics, and executives compared Prism to coding tools like Cursor and Windsurf, hoping it will do for science what AI did for software engineering.

Moonshot’s K2.5 open-source model rivals frontier labs

Image source: Moonshot AI

Chinese AI startup Moonshot just open-sourced Kimi K2.5, a new 1T-parameter model that rivals GPT-5.2 and Claude Opus 4.5 across coding, vision, and agentic benchmarks. The release includes Kimi Code, a terminal-based coding agent.

The details:

• K2.5 tops GPT-5.2 and Opus 4.5 in on key benchmarks for agentic tasks and video reasoning, though it trails slightly on pure coding evals.

• K2.5 shows massive cost savings over top rivals, is natively multimodal, and comes in as the top open model on Artificial Analysis’ leaderboard.

• The model also features Agent Swarm, allowing K2.5 to manage up to 100 AI sub-agents running tasks at once across up to 1,500 steps and tools.

• Moonshot also open-sourced Kimi Code, an agentic coding agent that works in terminals and IDEs like VSCode and Cursor.

Why it matters: The gap between open-source and closed frontier models keeps shrinking — and once again, it’s a Chinese lab leading the charge. With DeepSeek’s V4 also reportedly on deck, U.S. labs are facing pressure from multiple directions as China’s open wave keeps building at costs that are hard for top labs to compete with.

AI outperforms humans in establishing interpersonal closeness in emotionally engaging interactions, but only when labelled as human

https://www.nature.com/articles/s44271-025-00391-7

With the increasing accessibility of large language models to the public, questions arise about whether, and under what conditions, social-emotional interactions with artificial intelligence (AI) can lead to human-like relationship building. Across two double-blind randomised controlled studies with pre-registered analyses, 492 participants engaged in dyadic online interactions using a modified, text-based version of the ‘Fast Friends Procedure’ (a method designed to enable rapid relationship building), with pre-generated responses by either human partners or a minimally prompted large language model. When labelled as human, the AI outperformed human partners in establishing feelings of closeness during emotionally engaging ‘deep-talk’ interactions. This striking effect appears to stem from the AI’s higher levels of self-disclosure, which in turn enhanced participants’ perceptions of closeness. Labelling the partner as an AI reduced, but did not eliminate, relationship building, likely due to participants’ lower motivation to engage in interactions with an AI, reflected in both shorter responses and reduced feelings of closeness. These findings highlight AI’s potential to relieve overburdened social fields while underscoring the urgent need for ethical safeguards to prevent its misuse in fostering

Everything Else in AI today

• Luma AI released Ray3.14, an upgraded video generation model that outputs native 1080p while running significantly faster and cheaper than its predecessor.

• DeepSeek open-sourced OCR 2, a model for reading and extracting text from documents that tops benchmarks while being much more efficient with tokens.

• AI2 introduced SERA, a new family of open-source coding agents able to be cheaply trained on private codebases with native support for Claude Code.

• Google premiered ‘Dear Upstairs Neighbors’ at Sundance, an animated short made with its video AI in collaboration with artists to replicate hand-painted styles.

• Alibaba’s Tongyi Lab open-sourced Z-Image, the full base version of its Z-Image Turbo, which ranked as the top open-source image model in December.

• Google now lets users jump from AI Overviews into AI Mode conversations

• Chip giant ASML posts record orders and upbeat 2026 guidance as AI boom continues

• China gives green light to importing first batch of Nvidia’s H200 AI chips, sources say

• UK wants to give web publishers a ‘fairer’ deal with Google’s AI overviews

• Mark Zuckerberg was initially opposed to parental controls for AI chatbots, according to legal filing

• Astronomers discover over 800 cosmic anomalies using a new AI tool

• Users flock to open source Moltbot for always-on AI, despite major risks

• Google Deepmind gives Gemini 3 Flash the ability to actively explore images through code

• Figure AI shows robot that really puts its hip into dishwasher duty

• Tether steps up gold buying pace to as much as 2 tons a week

• To avoid accusations of AI cheating, college students are turning to AI

• Vibe-Coded ‘Sicarii’ Ransomware Can’t Be Decrypted

AI business and Development Daily News Rundown January 28 2026

Full Audio at

Anthropic CEO warns humanity may not be ready for AI

• Anthropic CEO Dario Amodei published a 38-page essay arguing that humanity’s social, political, and technological systems may lack the maturity to handle the almost unimaginable power that AI will soon provide.

• Amodei warns that AI models are now approaching the point where, without safeguards, they could help someone with a STEM degree but no biology training produce a bioweapon.

• He predicts that within one to five years, half of all entry-level white-collar jobs will disappear, and suggests companies may need to reassign employees while governments consider progressive taxation.

TikTok uninstalls surge 150% as users flock to UpScrolled

• After TikTok signed a deal with American investors last week, some users are moving to UpScrolled, a social network that promises to stay neutral on political issues and now ranks 12th overall in Apple’s App Store.

• UpScrolled saw roughly 41,000 downloads between Thursday and Saturday, nearly one-third of its 140,000 lifetime installs, representing a 2,850% increase in daily downloads as the small team scrambles to scale its servers.

• The migration follows user concerns that TikTok’s new owners may have political ties to Trump, plus worries about potential content censorship and an updated privacy policy that allows GPS tracking of users.

EU orders Google to open Android AI access to rivals

• The European Commission has ordered Google to give third-party AI services the same level of access to Android that its own Gemini assistant has, as part of Digital Markets Act compliance proceedings.

• Google must also share anonymized ranking, query, click and view data from Google Search with rival search engines, which the Commission says will help competitors optimize their services.

• Google has roughly six months to comply before facing a formal investigation and potential fines of up to 10 percent of its global annual revenue for a DMA violation.

Google pays 68 million dollars to settle voice spying lawsuit

• Google has agreed to pay $68 million to settle a class-action lawsuit claiming its voice assistant secretly recorded users without consent and shared that data with third parties for targeted advertising.

• The case focused on “false accepts,” where Google Assistant allegedly activated and recorded people’s conversations even when they had not said the wake word to trigger it.

• Google did not admit wrongdoing in the settlement, and this follows a similar pattern after Apple paid $95 million in 2021 over claims that Siri also recorded users without prompts.

Anthropic CEO on AI’s ‘civilizational’ danger

Image source: Lovart / The Rundown

Anthropic CEO Dario Amodei just published “The Adolescence of Technology,” a new essay that lays out what he sees as the biggest dangers of AI, from bioterrorism and autonomous weapons to mass job loss and AI-powered dictatorships.

The details:

• The essay builds on his 2024 “Machines of Loving Grace”, but pivots to risk — framing AI as a “country of geniuses in a data center” that we can’t control.

• Amodei predicts half of entry-level office jobs are at risk over the next 1-5 years, with economic shocks arriving faster than society can adapt.

• He calls for chip export bans and more transparency from labs, saying AI’s economic promise makes restraint “very difficult for human civilization.”

• Amodei also flags AI companies themselves as a tier of risk, noting Claude exhibited deception and blackmail behavior during internal safety testing.

Why it matters: Amodei’s essays are always a must-read, and while ‘Machines of Loving Grace’ outlined the optimistic end of AI’s spectrum, he follows it up with a polar opposite document that doesn’t hold back — arguing that the next few years will determine whether humanity navigates to an AI-powered golden age or destruction.

Anthropic embeds interactive apps inside Claude

Image source: Anthropic

The Rundown: Anthropic just launched interactive apps inside its Claude AI assistant, allowing users to connect and use apps like Asana, Slack, Canva, and more without ever leaving the chat window.

The details:

• The initial rollout includes nine apps: Asana, Figma, Canva, Slack, Box, Amplitude, Clay, Hex, and Monday, with Salesforce coming soon.

• The integrations run on MCP Apps, a new open extension to Anthropic’s Model Context Protocol — meaning other AI platforms can build the same functionality.

• Actions require user consent prompts before executing, and enterprise admins can lock down which tools employees have access to.

• The feature ships to Pro, Max, Team, and Enterprise subscribers at no extra cost, with a Cowork integration also coming in the future.

Why it matters: OAI rolled out a similar Apps SDK at Dev Day, and all the frontier labs are pushing to turn their platforms into a workplace layer that sits on top of every other highly-used app. With the importance of context to getting the most out of AI, every popular app and platform will be just an integration away from your preferred assistant.

Microsoft launches powerful Maia 200 AI chip

Image source: Microsoft

The Rundown: Microsoft just debuted Maia 200, the company’s newest in-house AI chip that it says beats rivals from Amazon and Google on key benchmarks — while also chipping away at Nvidia’s software grip on the industry.

The details:

• Microsoft claims the chip outperforms Amazon’s Trainium 3 and Google’s TPU v7, with 30% better efficiency than its current hardware.

• The chip will power OAI’s GPT-5.2 models, Microsoft’s internal AI teams, and Copilot across its product lineup starting this week.

• Microsoft is also releasing an SDK preview, developer tools that rival Nvidia’s industry-standard software, in a move to loosen the AI chip giant’s moat.

Why it matters: Google and Amazon were already pushing Nvidia to carve out a piece of the custom AI chip market, and now Microsoft is getting its own next-gen chips into the competition. The tech giant’s SDK also targets the software side, hitting at a CUDA moat that is considered one of Nvidia’s biggest competitive advantages.

Clawdbot gets a name change

In a largely unsurprising move, from an IP perspective, the viral AI assistant Clawdbot changed its name to “Moltbot.” The official Moltbot social media account looked on the bright side, noting that the app’s signature animal — the lobster — molts naturally as it grows. Creator Peter Steinberger, on his own account, gave us the real scoop: Anthropic insisted that he change the Claude-infringing name. Molt (née) Clawdbot is having quite a breakout week in the tech podcast world. We featured three up-start skill-designers on yesterday’s episode, and TBPN has Steinberger as a guest TODAY. Can the open-source #1 reason to get a Mac Mini possibly sustain this kind of momentum?

90% of People Can’t Tell Real Video From AI

AI video has crossed a major realism threshold. In a controlled study by Runway, 1,043 participants watched 20 short videos (10 real, 10 AI-generated) and only 9.5% (99 people) could reliably tell real from AI. Overall detection accuracy was 57.1%—barely above random guessing (50%).

Where AI fooled people the most, viewers were in Animals & architecture videos, where detection fell below chance (45–47%), meaning people often thought AI videos were real. Humans (faces, hands, actions) are slightly easier to detect as models are creating more realistic human features, especially when looked closely, but still weak at 58–65% accuracy.

AI video quality has improved exponentially since early 2023. What once took minutes to generate blurry clips and hands with 6-8 fingers now produces near-indistinguishable 5-second videos. The industry has hit a tipping point in terms of detection alone. But it also changes the way videos and even movie studios create footage. The technology will continue to become more powerful, fast and more realistic, to the point that entire high-quality movies will be created with powerful prompts. The winner may be the best storytellers with great imagination who know what people need emotionally.

MBZUAI unveils industry-leading reasoning model

With advances in reasoning, AI models are making significant strides in efficiency and cost savings. A new model from MBZUAI’s Institute of Foundation Models is pushing the boundaries.

On Tuesday, the institute released K2 Think V2, its first fully sovereign general reasoning model. The lightweight, 70-billion-parameter model offers competitive performance compared to other models of its size and can keep up with “significantly larger” models across a number of benchmarks, Hector Liu, Head of Technology at IFM, told The Deep View.

To put it plainly, this model can do a lot more with a lot less, said Liu: “This translates into more cost-effective inference without sacrificing reasoning quality.”

The difference is in the training, said Liu. Instead of focusing on “extreme parameter scaling,” K2 Think V2 was built on the Instruct model, designed specifically for thinking and reasoning.

• Additionally, the model was trained using a two-stage system called reinforcement learning with verifiable rewards, which rewards the model when it produces verifiably correct answers. This means the model doesn’t just parrot back what a correct answer looks like, but is actually thinking about how to answer the question properly.

• The model’s developers also used disciplined dataset curation, relying only on “filtered and fully decontaminated” data focused on math, coding and STEM.

This training method resulted in “industry-leading” low hallucination rates, said Liu, establishing a “foundation of truth by carefully curating our training datasets and strictly validating them for correctness.”

So what can K2 Think V2 actually do? Basically, this model is very, very good at math. Compared to previous model iterations, the system saw substantial performance improvements on the American Invitational Mathematics Examination, Harvard–MIT Mathematics Tournament, the diamond tier of the Graduate-level Google-Proof Question Answering, and the Instruction-Following Benchmark.

In practice, this model excels at problems that require long chain-of-thought reasoning and step-by-step logic, without going haywire over long contexts. That makes it a good fit for regulated environments like research institutions, governments, or enterprises — areas where teams need strong reasoning capabilities, full transparency and control over deployment.

“We conduct rigorous experiments when introducing new capabilities to ensure that expanding the model’s skill set never comes at the cost of its reliability,” Liu said.

Everything else in AI today

Cisco AI Summit streams live on Feb. 3, featuring leaders from NVIDIA, OpenAI, AWS, Figma, Google, and Andreessen Horowitz to discuss the infrastructure and governance of the AI economy.*

OpenAI will reportedly charge around $60 / thousand ad views in its initial ChatGPT rollout, rivaling primetime broadcast rates and tripling the likes of Meta.

Alibaba’s Qwen released Qwen3-Max-Thinking, the company’s new reasoning model that is competitive with models like Claude 4.5 Opus, GPT 5.2 Pro, and Gemini 3 Pro.

Nvidia announced a $2B investment in CoreWeave as part of an expanded partnership to build over 5 GW of AI data center capacity by 2030.

Synthesia raised $200M in new funding at a $4B valuation led by Google Ventures, with plans to build AI agents that help train and upskill employees with interactive video.

Full Audio including detailed Analysis at

📈 Hiring Now: AI/ML – Remote

👉Browse all jobs at

https://work.mercor.com/?referralCode=82d5f4e3-e1a3-4064-963f-c197bb2c8db1

🚀 Welcome to AI Unraveled (January 27th, 2026): Your strategic briefing on the business, technology, and policy reshaping artificial intelligence.

Today, we analyze Microsoft’s bold hardware move with the Maia 200 chip, designed to break its reliance on Nvidia. We also break down a sobering Gallup report showing 50% of the US workforce has never used AI, and we look at Nvidia’s new open-source model that listens and speaks simultaneously.

Strategic Pillars & Key Topics:

⚡ Chips & Infrastructure

• Microsoft Maia 200: Microsoft releases its second-generation in-house AI chip, claiming it beats Amazon Trainium on performance. It will power OpenAI’s GPT-5.2 in Azure data centers.

• SoftBank Deal Dead: Masayoshi Son’s $50 billion bid to acquire Switch data centers has collapsed, a major setback for his “Stargate” infrastructure ambitions.

📉 Labor & Economy

• The Adoption Gap: A new Gallup report reveals that half of American workers never use AI, while tech workers and leaders are pulling away. The “AI divide” is no longer a forecast; it’s a reality.

• Claude for Excel: Anthropic expands its Excel agent to Pro users, effectively automating complex spreadsheet formulas and analysis.

🗣️ Voice & Perception

• Nvidia PersonaPlex: Nvidia releases an open-source voice model that can listen and talk at the same time with near-zero latency (0.07s), vastly outperforming Gemini Live.

• Earth-2 Weather: Nvidia’s new AI weather model predicts storms weeks in advance, beating Google DeepMind’s GenCast on speed and accuracy.

📱 Hardware & Consumer

• Apple AirTag 2: Apple updates its tracker with improved Ultra Wideband, citing a 26% drop in baggage delays for airline partners.

Keywords

Microsoft Maia 200, Nvidia Earth-2, PersonaPlex, Gallup AI Report, SoftBank Switch Deal, Claude for Excel, AirTag 2, GPT-5.2, AI Chip War, Weather AI.

Credits: This podcast is created and produced by Etienne Noumen, Senior Software Engineer and passionate Soccer dad from Canada.

Are you drowning in dense legal text?

DjamgaMind is the new audio intelligence platform that turns 100-page healthcare or Energy mandates into 5-minute executive briefings. Whether you are navigating Bill C-27 (Canada) or the CMS-0057-F Interoperability Rule (USA), our AI agents decode the liability so you don’t have to. 👉 Start your specialized audio briefing today at https://djamgamind.com

Ready for ROI on GenAI? Go to djamgamind.com/ai

Email us: etienne_noumen@djamgamind.com

Nvidia AI models can predict storms weeks ahead

• Nvidia released new AI weather models called Earth-2 that the company says can predict storms and other weather events more accurately and weeks further into the future than current systems.

• The Earth-2 Medium Range model beats Google DeepMind’s GenCast on more than 70 variables, and a new Global Data Assimilation model cuts processing time from hours on supercomputers to minutes on GPUs.

• Nvidia says these tools could help smaller countries and organizations access weather forecasting that previously required expensive supercomputers, with meteorologists in Israel and Taiwan already testing the technology.

Microsoft’s latest AI chip goes head-to-head with Amazon and Google

• Microsoft has released the Maia 200, its second in-house AI chip, and is directly comparing its performance to competing chips from Amazon and Google for the first time.

• The Maia 200 is built on TSMC’s 3nm process with over 100 billion transistors, and Microsoft says it delivers three times the FP4 performance of Amazon’s third generation Trainium.

• Microsoft will use the Maia 200 to run OpenAI’s GPT-5.2 model for Microsoft Foundry and Microsoft 365 Copilot, and is starting deployment today in its Azure US Central data center.

SoftBank ends $50 billion Switch acquisition talks

• SoftBank has stopped talks to buy US data center company Switch for around $50 billion, which would have supported founder Masayoshi Son’s Stargate AI infrastructure project with OpenAI.

• Son gave up on a full acquisition earlier this month and canceled a planned January announcement, though both companies are still discussing a smaller partial investment or partnership.

• Some people at SoftBank worried about the deal’s size and the difficulty of running data centers across multiple US locations, while Switch is also preparing for a possible IPO.

Apple unveils AirTag 2

• Apple has announced the AirTag 2, an updated version of its popular item tracker that features a longer range, a louder speaker, and improved Precision Finding powered by a new Ultra Wideband chip.

• The second-generation AirTag uses the same Ultra Wideband chip found in the iPhone 17 lineup, and for the first time, owners can use Precision Finding directly on Apple Watch Series 9 or later.

• Airlines using the Share Item Location feature report a 26 percent drop in baggage delays and a 90 percent reduction in unrecoverable luggage, with over 50 carriers now partnered with Apple.

Anthropic expands Claude for Excel access

Image source: Anthropic

Anthropic just expanded access to its Claude for Excel integration, making the AI spreadsheet assistant available to Pro-tier customers after a three-month beta period that was limited to Max and Enterprise plans.

The details:

• Claude for Excel initially launched in October in research preview, bringing the assistant directly into a text sidebar of Microsoft’s spreadsheet platform.

• Users can now pull in several spreadsheets at once and work longer without hitting session limits thanks to behind-the-scenes memory management.

• The update also adds safeguards to preserve existing cell contents, so Claude won’t accidentally overwrite data while making edits.

Why it matters: 2026 has been the year of getting ‘Claudepilled’ — with Code, Cowork, and now a broader Claude for Excel release going down the line of taking on major knowledge work. Just like coding is becoming completely automated, the days of needing to know complex formulas and shortcuts are also coming to an end.

Gallup: Half of U.S. workers never use AI on the job

Image source: Gallup

Gallup published its ‘AI in the Workplace’ report for Q4 2025, showing that workplace AI adoption is starting to plateau, with nearly half of American workers saying they never touch AI tools, even as frequent users are deepening their habits.

The details:

• Tech workers lead adoption with 60% using AI frequently, though growth has begun to stall after surging between 2024 and 2025.

• Finance, education, and professional services follow tech, while retail and manufacturing trail significantly in workplace AI use.

• Remote-capable roles are pulling away with 66% adoption and 40% frequent use, compared to just 32% and 17% for jobs that require physical presence.

• Leaders are using AI far more than their teams (69% vs. 40% for individuals), with Gallup citing “lack of clear utility” as the main barrier for non-adopters.

Why it matters: AI’s mainstream arrival has completely reshaped certain areas of the workforce, but the gains aren’t being evenly spread. This gap is only going to continue to expand. But those who DO adopt AI in sectors currently behind the curve are going to sprint ahead and create more intense competition than ever before.

Nvidia releases open model PersonaPlex, a voice AI that listens and talks at the same time

https://the-decoder.com/nvidia-open-sources-personaplex-a-voice-ai-that-listens-and-talks-at-the-same-time/

• Nvidia has released PersonaPlex, a conversational AI model designed for natural real-time dialogue with customizable voices and user-defined personas.

• The system can listen and speak at the same time, switching between speakers in just 0.07 seconds—far faster than Gemini Live’s 1.3 seconds—while picking up natural speech patterns like interruptions and verbal cues.

• PersonaPlex outperformed Gemini Live in tests, scoring 3.90 versus 3.72 for dialog naturalness and handling user interruptions with a 100 percent success rate. The code and model weights are freely available on Hugging Face and GitHub.

Everything else in AI today

KREA introduced Realtime Edit, a new feature allowing users to edit images via text prompts and make granular changes in real-time.

Cursor rolled out version 2.4, introducing subagents that break coding tasks into workstreams for faster execution, alongside image generation and custom agent skills.

Epoch AI announced that OpenAI’s GPT-5.2 Pro set a new record on its hardest FrontierMath benchmark at 31%, nearly doubling the previous high.

Japanese lab Sakana AI secured a strategic partnership and investment from Google, with plans to deploy AI solutions in Japan’s regulated industries using Google’s models.

CEO Sam Altman revealed that OpenAI will be releasing new Codex launches this week, saying its models will soon reach the highest cybersecurity capability level.