Claude produced formal proofs for code from 123 OSS-Fuzz libraries
Original title: Anthropic's Claude now writes code and proves it safe automatically
Rocq checks the proofs, which hold only for one architecture, single-threaded code, and inputs under 64 MB.
Anthropic's Frontier Red Team and partner Theorem used Claude to generate formal proofs for existing code drawn from 123 open-source libraries in Google's OSS-Fuzz project. The model produced about 180 million lines of proof in a few weeks, and Rocq checks those proofs. While proving, Claude found and fixed hundreds of bugs, including security bugs in a widely used PDF reader. Anthropic says the proofs apply only under a specific processor architecture, to single-threaded programs, and to inputs under 64 MB.
Unconfirmed: Anthropic does not report total project cost, and the proofs do not cover other hardware, multithreading, or larger inputs.
- The article cites 20 person-years to formally verify the seL4 kernel.
- One week on a PDF reader matched a typical year of reported security bugs.
- Claude found a logic error that had gone undetected for over a decade.
- Opus 5.5 used about 40 percent fewer tokens than Opus 5 on half the proofs.
- Theorem's roadmap names glibc, OpenSSL, and cURL.
From the original article
2 min read at the-decoder.com
Anthropic had Claude generate mathematical proofs that existing code is correct, producing roughly 180 million lines of proof code for 123 open-source libraries. Along the way, the model caught bugs that years of testing had missed. Anthropic's goal: making formal verification with AI as routine as standard testing.
By Matthias Bastian · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.