OpenAI cracks a 90-year-old math problem, Meta launches Muse AI agent, and Mistral raises €3B

A 90-year-old math problem. 10,000 AI agents. 88 hours. And now a controversy that may overshadow the result entirely.

The Navier-Stokes problem that OpenAI claims to have solved is one of the seven Millennium Prize Problems. How much prize money does the Clay Mathematics Institute award for solving each one?

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

OpenAI announced this week that a fleet of 10,000 AI agents - running on an internal model described as more powerful than the recently released GPT-6 Astra - cracked the Navier-Stokes existence and smoothness problem in just 88 hours. It's one of seven Millennium Prize Problems, some of the most important unsolved questions in mathematics, and a $1 million bounty has sat unclaimed for decades.

But the announcement has been clouded almost immediately by accusations of impropriety. According to MIT Technology Review, the claim has been "overshadowed by accusations" - and TechCrunch reports that an NYU mathematician says OpenAI "fought dirty" on what was a career-defining problem for researchers in the field. The tension sits right at the intersection of AI's hunger for milestones and academia's careful process for verifying them.

The Navier-Stokes problem relates to the mathematical description of fluid and gas flow - think turbulence in aircraft engines or ocean currents. Solving it would have enormous implications for physics and engineering. The controversy isn't just about credit, though - it also raises a deeper question about what it means for AI to "solve" a math problem, and whether the result is verifiable the way a human proof would be.

If the result holds up, it's one of the most significant AI achievements ever - but the controversy around how OpenAI handled the announcement may define this story just as much as the math itself.

Read the full story (MIT Technology Review) → | Read the full story (The Guardian) →

Meta launched Muse, a personal AI agent designed to compete directly with rivals like OpenClaw and Instinct. According to Wired, Muse can do everything from sell your car to book you a plane ticket. But to do any of that, it needs access to your email, calendar, payments, health services, and more - making it Meta's biggest consumer AI product to date.

Meta says privacy is built into Muse by design. The company frames this as its most important step yet in a multi-billion-dollar strategy to catch up to OpenAI, Anthropic, and Google in the AI race. The Verge notes it's positioned as putting "AI in the hands of virtually anyone."

Here's the tension that TechCrunch puts front and center: will consumers actually hand Meta this level of access? The company has spent years rebuilding its reputation on data privacy, and Muse is essentially asking users to go all-in. It's a genuinely useful product concept - a single agent that can manage your life across services - but the name on the door matters enormously for adoption.

Muse is the most coherent AI agent pitch Meta has made yet - but its success depends entirely on whether people trust Meta more than they did five years ago.

Read the full story →

French AI lab Mistral AI just closed a €3 billion Series D at a €21 billion valuation, led by Samsung, Scaleup Europe, and PSG Equity. The raise cements Mistral's position as Europe's flagship AI company - and the timing is no accident. Demand for AI infrastructure that sits outside US jurisdiction is accelerating among governments and enterprises that don't want their data or models dependent on American hyperscalers.

TechCrunch frames this squarely in the "sovereign AI" trend - the idea that nations and regions need their own AI capabilities rather than relying entirely on US labs. Mistral has leaned hard into this positioning, building partnerships with European governments and offering models that can be deployed on local infrastructure.

At €21B, Mistral is now valued as a serious long-term rival to the US labs - and Samsung's involvement signals that Asian enterprises are just as interested in non-US AI options as European governments are.

Read the full story →

Google DeepMind unveiled AlphaGenome Atlas this week - described by The Verge as a "predictive map of every possible DNA letter change in the human genome." The platform is designed to help researchers understand how genetic mutations affect biology, with the goal of accelerating drug discovery and opening paths to new treatments for disease.

This follows the same playbook as AlphaFold, which transformed protein structure prediction. DeepMind's researchers say AlphaGenome Atlas could similarly transform genomics research - shifting the field from "what mutations exist" to "what each mutation actually does." That's a meaningful distinction for anyone working on genetic disease, cancer research, or precision medicine.

If AlphaGenome Atlas delivers on its promise at scale, it could do for genomics what AlphaFold did for proteins - compress decades of experimental biology into a searchable, queryable AI tool.

Read the full story →

Two separate Anthropic stories dropped this week and neither is good news for the company. First, TechCrunch reports that hackers are actively stealing Claude API tokens from subscribers - one user noticed his account draining tokens while he wasn't even working. Anthropic has since warned users directly about the attack vector.

Separately, The Verge reports that a group of Claude power users filed an expanded class action lawsuit today, alleging that Anthropic "deceptively" misled them into believing top-tier subscription plans would deliver more than they actually did. Anthropic has publicly prioritized power users - even cutting off other popular apps to preserve capacity for them - which makes the lawsuit particularly awkward.

If you're a Claude subscriber, check your token usage carefully - and if you're evaluating enterprise AI subscriptions, today's class action is worth reading before you commit to annual billing.

Read the token theft story → | Read the lawsuit story →

Quick side note: if you've been thinking about building a website for a project or portfolio, 60sec.site is an AI website builder that does exactly what the name suggests - worth a look if you want something live fast without touching code. And for more daily AI coverage, you can always browse the full archive at dailyinference.com.

The answer is $1,000,000! The Clay Mathematics Institute offers a $1 million prize for each of the seven Millennium Prize Problems. Only one has ever been solved before - the Poincaré Conjecture, by Grigori Perelman in 2003 (who famously declined the money). If OpenAI's Navier-Stokes solution holds up under scrutiny, it would be only the second in the set to fall.

Meta's Muse agent wants access to your email, calendar, payments, and health data. Would you actually hand that over - and does it matter which company is asking? Hit reply and let me know your honest answer. I read every response!

That's it for today - see you tomorrow with more. And if someone forwarded this to you, you can subscribe at dailyinference.com.