← Latest AI News

November 9, 2026 · Daily brief

AI proved an 18-year math problem in three days

Sovereignty angle
If AI can crack expert-level research in three days, your small team can now hand it work that used to mean hiring a specialist: legal research, financial modeling, technical writing, patent searches.

A startup used an unreleased Anthropic model to prove the Aaronson-Ambainis conjecture, a quantum computing problem open since 2008, verified by Lean.

What changed

Physical Superintelligence, a research lab, announced November 6 that an AI system solved the Aaronson-Ambainis conjecture. The problem, open since 2008, asks when quantum computers can beat regular computers by a huge margin.

The system used an unreleased Anthropic model. It produced a complete proof on day three. A tool called Lean checked every logical step and accepted it. Four Stanford researchers who solved a smaller version in August are now reading the full proof.

Lean works like spellcheck for math. You write the proof in strict format. Lean confirms every step follows or rejects it. A confident AI can't charm its way past.

What you can do with it now

You can use today's public AI models — ChatGPT, Claude, Gemini — for work that used to require expert help. Try legal research for a contract dispute. Ask it to model three pricing scenarios with your actual revenue numbers. Have it write a technical specification for a feature your developer needs to build.

The output won't be final. But it will be specific enough that you can read it, mark what's wrong, and ask the AI to fix those parts. That turns a $3,000 specialist job into a $300 review job.

For tasks with right and wrong answers — like the math proof — test the AI on a problem you already know the solution to. If it gets that right, try it on the real problem. If it gets it wrong, you've learned the task is still outside its range.

What to keep in your own hands

Never use AI output as a final answer when the stakes are legal, financial, or medical. The math proof worked because Lean checked every step. You don't have Lean for a contract or a tax filing.

Always have a person who knows the subject read the AI's work before you act on it. If you're using AI for legal research, have a lawyer review it. If you're using it for financial modeling, have your accountant check the formulas.

The Anthropic model that solved this problem isn't public yet. When it is, it will likely cost more than the current models. Don't assume the breakthrough means your $20 subscription suddenly got smarter. You'll need to upgrade and pay more to access this level of capability.