Google’s AI agent teams produce machine-checked proofs of hard math

Google Research described /teamwork, a feature in its Antigravity tool that runs many AI agents in parallel and checks their work with strict verifiers such…

November 13, 2026 · Science

Google Research described /teamwork, a feature in its Antigravity tool that runs many AI agents in parallel and checks their work with strict verifiers such as the Lean 4 proof checker. Google says the system produced a fully machine-checked version of the Weil conjectures, about 1.8 million lines of Lean, and a formalization of a recent proof on Hilbert’s tenth problem of more than 919,000 lines. It also says a team used it to prove the Courtade–Kumar conjecture, an information theory problem from 2013, and to formalize that proof. Google notes the work was done with human experts, who must still confirm that each formal statement matches the original problem. Outside researchers can use /teamwork in Antigravity now, and Google is offering 10,000 academic researchers six months of free top-tier access.

Why it matters to you: A proof that a computer has checked line by line does not depend on trusting the AI that wrote it. That is the pattern to watch: AI doing long, hard work with a strict checker behind it. These are Google’s own figures and have not been independently reviewed.

Source: Google Research

More from Today in AI

All of Today in AI. Where AI is heading.

Related Power of AI pages

Keep reading with Start here, everyday uses, the tool guide, the writing workshop, and sources and standards.