[AI DAILY NEWS RUNDOWN] AI Solves 56-Year-Old Math, Sanctions Backfire, and the Workforce Lie (May 25 2026)

🎧 Listen ADS-FREE: https://podcasts.apple.com/us/channel/djamgamind/id6760446113

Visit our Research Hub at https://djamgamind.com

Subscribe to our Newsletter at https://enoumen.substack.com

#DJAMGAMIND

Summary: In today’s briefing, we analyze “The Geopolitical Backfire and The Death of the Turing Test.” We deconstruct the monumental milestone of Google DeepMind’s AlphaProof Nexus solving nine open Erdős math problems, marking a shift to original, machine-speed discovery. We explore the failure of US hardware blockades as Chinese lab DeepSeek slashes prices by 75% using homegrown Huawei chips. We dive into the severe corporate morale crisis, with new data showing 63% of workers are actively lying about their AI skills due to automation anxiety. We also cover Anthropic’s Claude Mythos finding 10,000+ critical bugs, Pope Leo XIV’s 42,000-word AI encyclical, and Waymo robotaxis getting defeated by puddles.

Today’s Sponsor:

• 🛑AIRIA: The ultimate zero-trust AI security layer. Deploy autonomous agents safely without compromising your enterprise data. 👉 Govern your agents: https://airia.com/request-demo/?utm_source=AI+Unraveled+&utm_medium=Podcast&utm_campaign=Q1+2026

Important Topics:

• Google AI Cracks Unsolved Math: Google DeepMind’s AlphaProof Nexus generates machine-verified proofs for nine open Erdős problems, including two unsolved for 56 years.

• The Geopolitical Blockade Backfires: Chinese lab DeepSeek slashes V4-Pro prices by 75% ($0.435 per million tokens) operating on Huawei’s Ascend supernodes, severely undercutting US frontier models.

• The Workforce AI Lie: A GCheck study finds 63% of workers (and 80% of Gen Z) exaggerate their AI skills out of job insecurity, creating a “double distortion” where they fake proficiency while actively limiting AI use.

• Claude Mythos Finds 10K+ Bugs: Anthropic’s Project Glasswing reveals that the unreleased Mythos Preview model helped partners find over 10,000 high- or critical-severity vulnerabilities in just one month.

• Pope Leo XIV’s AI Encyclical: The Pope releases “Magnifica Humanitas,” a 42,300-word document calling to “disarm AI” and warning that the technology weakens personal judgment and genuine human connection.

• Waymo Robotaxis Halted by Water: Waymo pauses service in four major US cities after its robotaxis struggle and freeze when attempting to cross flooded intersections.

• China Assigns Digital IDs to Robots: China rolls out a national program assigning a unique, 17-digit digital identity code to every humanoid robot manufactured in the country.

• Google CEO Responds to Boos: Sundar Pichai addresses the wave of graduates booing AI at commencements, acknowledging that workforce anxiety is “rightfully” growing as automation accelerates.

🔗 RESOURCES

Stop scrolling through generic “Top 10” lists and head over to the AI Executive Toolkit at https://djamgamind.com/toolkit

• Find REMOTE AI Jobs (Mercor): Apply Here

• Google Workspace: Professionalize your firm’s infrastructure with secure, cloud-based collaboration and branded communication: https://referworkspace.app.goo.gl/Q371

⚗️ PRODUCTION NOTE: We Practice What We Preach.

AI Unraveled is produced using a hybrid “Human-in-the-Loop” workflow.

Google’s AI cracks nine unsolved math problems

Image source: Images 2.0 / The Rundown

The Rundown: Google DeepMind’s AlphaProof Nexus, an AI system that generates machine-verified mathematical proofs, solved nine open Erdős problems, including two unsolved for 56 years, just a day after OpenAI claimed its own Erdős breakthrough.

The details:

• The system paired an LLM with Lean, a proof assistant, to generate machine-verified proofs for the nine problems spanning combinatorics and graph theory.

• Each problem cost a few hundred dollars to solve, with the AI also proving 44 open conjectures from the Online Encyclopedia of Integer Sequences.

• A simpler version of the agent matched the results but cost more, and problems requiring new mathematical constructions remained out of reach.

• OpenAI’s win last week saw its AI disprove an 80-year-old Erdős conjecture — months after walking back a claim of solving 10 novel problems.

Why it matters: Google’s progress on math problems unsolved for decades shows how fast AI is moving toward original solutions, and how formal verification changes the game. The system generates proofs, verifies them in Lean, and repeats until one passes. Over time, this will help researchers make novel discoveries at machine speed.

Claude Mythos finds 10,000+ critical vulnerabilities

Image source: Anthropic

The Rundown: Anthropic shared the first results from Project Glasswing, revealing that Claude Mythos Preview and its ~50 partners have found 10,000+ high- or critical-severity vulnerabilities in just one month.

The details:

• Cloudflare alone found 2K bugs with a false positive rate better than human testers. Mozilla found and fixed 271 vulnerabilities in Firefox 150.

• Anthropic also scanned 1,000+ open-source projects, with Mythos flagging 6,202 as high/critical. After independent triage, 62% (or nearly 3,900) held up.

• Mythos detection went beyond vulnerability flagging, with one partner bank using Mythos to detect and block a $1.5M fraudulent wire transfer.

• Now, Glasswing will expand to additional partners, including U.S. and allied governments, with a general release of Mythos-class models to follow.

Why it matters: Anthropic says Mythos remains gated because no company — including itself — has safeguards strong enough to prevent misuse. But with OpenAI ramping up its cyber models and Chinese players catching up, equally capable (or better) AI will emerge. When it does, how fast the world can patch will be the real test.

The Blockade Backfired

What’s happening: DeepSeek, the Chinese AI startup known for frontier models at fractional prices, made permanent a 75 percent cut on V4-Pro. Tokens now run $0.0035 to $0.83 per million, far below any US frontier model. V4-Pro had been priced 12 times its lighter sibling on “compute capacity constraints,” constraints that broke with Huawei’s Ascend 950 supernodes ramping in H2.

How this hits reality: The export-control bet was that cutting off Nvidia would choke China’s compute and hold the AI frontier for US labs. All three pieces broke. Huawei shipped Ascend, DeepSeek built V4-Pro on it, prices fell 75 percent. Sanctions didn’t slow Chinese AI. They subsidized it. DeepSeek now sells frontier-tier AI at a fraction of US prices, on the chip Washington banned. Anthropic and OpenAI bet developers pay premium for the best; DeepSeek bets most pay for good-enough at near zero. Keep going, and the frontier stops being where developers buy.

Key takeaway: Washington banned the chip to slow Chinese AI, China built one anyway, and the cheapest frontier AI on the market now runs on it.

Pope Leo releases manifesto on AI LINK

• Pope Leo XIV has issued his first encyclical, Magnifica Humanitas, a 42,300-word document on artificial intelligence that calls for regulation of the technology and a moral framework to protect humanity for generations to come.

• The Pope urges readers to “disarm AI,” which he defines as freeing technology from monopolistic control and opening it to debate, while also cautioning against deploying AI in warfare and the workplace.

• He warns that AI’s speed can weaken personal creativity and judgment, and that its imitation of human communication risks making users lose the desire to form genuine human connections with real people.

Huawei targets 1.4nm chips by 2031 LINK

• Huawei will roll out its next-generation Kirin smartphone chips later this fall, featuring a new design framework called “LogicFolding” that the company developed without access to ASML’s extreme ultraviolet lithography machines.

• Instead of shrinking transistors further, LogicFolding stacks multiple layers of circuitry on a single chip and reorganizes how processing elements talk to each other, aiming for performance gains through better data flow.

• Huawei says it has spent six years refining these techniques under US export controls and has mass-produced 381 chip models, targeting transistor densities comparable to a 1.4-nanometer process by 2031.

China assigns digital IDs to humanoid robots LINK

• China has rolled out a national program that gives every humanoid robot made in the country a unique digital identity code, working like a citizen ID for two-legged machines that can walk and run.

• Each code has four parts: a two-digit national code for tracking shipments, a four-digit manufacturer code, a six-digit product model code, and a 17-digit serial code identifying individual units from production to recycling.

• More than 28,000 robots across 200 models have already received a digital ID before the public announcement, with the rules covering manufacturers, service providers, sellers, end users, and recycling facilities across the supply chain.

Google CEO Sundar Pichai responds to graduates booing AI LINK

• Google CEO Sundar Pichai has responded to the wave of graduates booing AI at recent commencement ceremonies, saying on the Hard Fork podcast that even skeptical students will help shape the technology’s future direction.

• Pichai’s remarks, made ahead of his own Stanford commencement speech, followed former Google CEO Eric Schmidt being loudly booed at the University of Arizona after he praised AI’s potential and compared it to major industrial revolutions.

• Pichai said anxiety about AI is “rightfully” growing, pointing to a workforce reshaped by automation, with Business Insider reporting unemployment among recent graduates has hit a four-year high as companies adopt more AI tools.

Tether launches official stablecoin in Georgia LINK

• Tether, the largest stablecoin issuer, said it will roll out GELT, a crypto token tied to the Georgian lari, with backing from Georgia’s government as the country leans into digital assets.

• Tether described GELT as a “digital representation of the Georgian lari” meant to support cross-border commerce, fintech, and digital payments, though details on its structure, rollout, and implementation are still to come.

• The company credited the National Bank of Georgia’s stablecoin rules for drawing it in, and noted Georgia, home to 3.7 million people, ranks among the world’s top miners of cryptocurrency.

Report: Workers are lying about their AI skills

Though enterprise executives are eager to ramp up their AI efforts, the tech may be causing a workforce morale problem.

A study published this week from GCheck found that, faced with anxiety about how automation could impact job security, 63% of 1,500 workers surveyed reported that they exaggerate their AI skills to appear more up-to-date. That number shot up to 80% among Gen Z workers as the tech threatens early-career and entry-level roles more drastically.

The GCheck report found that AI is pulling on workers from both ends, leaving them caught between the fear that AI will disrupt their jobs and the pressure to appear as AI power users.

• Nearly 70% of workers reported that they believe AI will automate part of their responsibilities. And these concerns may not be unfounded: 40% observed that AI tools are already, in part, doing their jobs.

• Only 38% said they feel prepared to use AI tools effectively, while 22% said they would struggle to use AI or wouldn’t be able to use it at all.

• Still, many aren’t voicing their concerns. Around 40% said they speak confidently about AI in meetings to avoid appearing behind, and 33% let others assume that they have strong AI skills.

• A quarter have taken full credit for AI-assisted work, and 16% admitted to outright lying about their AI skills. Largely, employees are doing this due to pressure to appear AI-capable, fear of losing their jobs, or feeling they don’t have a choice.

However, despite talking up AI, the impact on morale is influencing worker behavior towards the tech, GCheck CEO Houman Akhavan told The Deep View. Around 81% of those surveyed admitted that they discourage or limit using AI at work. Akhavan called this contrast “double distortion.” Rather than relying on AI tools, workers are choosing manual work to avoid them.

“[It’s] a workforce overstating its AI abilities and quietly undermining AI adoption in the same organizations,” Akhavan said.

To combat this, enterprise leaders need to approach these anxious workforces with more empathy, Akhavan said. Workers need to feel safe enough to be candid about what they know and do not know, without the fear of job loss looming over them. More than half of the workers who reported exaggerating their AI skills never actually received any training. “That is a learning culture problem dressed up as a technology problem,” Akhavan said.

Robotaxi reality check: Where humans still win

There are times to take a robotaxi, and there are times when you need a human driver. I found that out the hard way this week.

It all started with my bright idea on Tuesday morning. I was staying in San Francisco and needed an early ride down to Shoreline Amphitheatre in Mountain View for the Google I/O keynote. “Wouldn’t it be fitting to take a Waymo to the event?” I thought. “I’ll just leave plenty early.” Big mistake.

I wanted to check into Google I/O when the doors opened around 8:00 a.m. to get a good seat for taking photos. So I planned to leave around 7:15 a.m. since the ride would take about 45 minutes in an Uber. Since a Waymo usually takes a little longer to hail, and it doesn’t drive as aggressively, I figured it would take a little extra time. Plus, there might be extra traffic.

So I wasn’t surprised when the Waymo picked me up at 7:20 a.m.and calculated an estimated arrival time of 8:32 a.m. Sure, that sounded a little slow, but I didn’t sweat it since seating at Google’s keynote wouldn’t begin until 8:30 a.m. anyway.

The trip went smoothly from pickup at my hotel in Mission Bay to getting on Highway 1 and cruising for about 30 miles to getting off on Exit 400A and getting onto Amphitheatre Parkway, about a mile from the venue. We were still on track for an on-time dropoff. And then things fell apart.

The traffic in Mountain View was completely backed up and was being routed in very specific ways. At one point, the Waymo waited very patiently in a line of cars that didn’t move for about 10 minutes. An experienced Uber driver would likely have switched lanes and found a different route. After finally moving through the traffic snarl, the Waymo then did something unexpected.

It turned left and went onto a back road where few other cars were going. That ended up putting us in the parking lot of a Google campus building. Once there, the Waymo started acting as if it was uncertain which way to turn. So, I tapped the Support button on the backseat screen, and a Waymo agent came on the speakerphone to help. As we talked through the situation, the Waymo eventually circled around the perimeter of the parking lot and pulled onto a street where it could take me to the drop-off point in Lot B of the Shoreline Amphitheatre.

The Waymo dropped me off at 9:09, 37 minutes after the originally calculated drop-off time and almost two hours after picking me up. The ride cost $102.72, and afterward, Waymo sent me a $10 credit to apply to a future ride, for my troubles. Luckily, Sabrina Ortiz saved me a seat at the Google I/O keynote, and all ended well.

Based on this experience, I’ve come up with a list of situations where I’ll avoid Waymo (and other robotaxi services) in the future and take an Uber instead:

• When I’m in a hurry and have a really important meeting or event, and it’s critical that I be on time

• When there could be irregular traffic situations and road closures

• When there are unnamed roads and places that could put the vehicle in unfamiliar territory

• When there is the potential for heavy traffic that may demand the ability to think creatively about finding alternative routes

Robotaxi Can’t “Cross” Water

What’s happening: Waymo paused service in Atlanta, San Antonio, Dallas, and Houston this week after a robotaxi drove into a flooded Atlanta intersection and sat stuck for an hour. The pauses followed last week’s NHTSA recall over flood-handling Waymo admits has no final remedy.

How this hits reality: Self-driving’s wall now look like isn’t regulation, geofencing, or compute. It’s water. Tesla bet on cameras only, with Musk calling LiDAR a crutch, and the cars once reportedly rushed into lakes recently. Waymo bet on the full stack and the cars freeze in the puddle. After tens of billions and a decade, the industry that promised to outdrive humans has share the one wall that love water. The frontier of autonomous driving may not general intelligence anymore. It’s whoever ships the puddle patch first wins the city.

Key takeaway: Self-driving cleared every benchmark engineers wrote for it and got stopped by the one humans have known about since they invented shoes: don’t step in the puddle.

What Else Happened in AI on May 25th 2026?

DeepSeek permanently cut V4-Pro pricing by 75%, bringing it down to $0.435 per million input tokens and $0.87 per million output tokens, far below closed-source rivals.

Perplexity open-sourced Bumblebee, a scanner for macOS and Linux that checks for risky packages, extensions, and AI tool configs during supply-chain incidents.

NVIDIA released NV-Generate-MR-Brain, a foundation model that generates synthetic 3D brain MRI scans and annotations, to accelerate medical imaging AI development.

McKinsey is rethinking its billing model as AI reduces the value of billable hours and clients demand fees tied to business outcomes, FT reported.

The White House approved $9B to help U.S. spy agencies acquire advanced AI chips amid concerns they’re falling behind in deploying frontier models, the NYT reported.

Starbucks scrapped its AI inventory system after nine months, citing persistent miscounts and mislabeled products across North American stores.