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.