Anthropic and Theorem used Claude to patch and verify 239 fuzzing entry points across 123 open-source libraries, proving that they do not crash for inputs smaller than 64 MB. A proof checker validates roughly 180 million lines of Rocq code. The guarantees assume single-threaded, statically linked RISC-V binaries with fixed memory budgets and a Linux execution model; they do not establish that entire libraries are bug-free. The separate Vero evaluation likewise distinguishes passing individual specifications from completing every proof in a repository, underscoring the importance of checking proof scope and trusted foundations.
Back to Tech
Tech
Claude and Theorem verify entry points in 123 libraries
Rocq checks the proofs, but bugs outside the specified properties and execution model remain possible.