AI News Page

← 2026-10-28 edition

THE SHORT VERSION

CLAUDE PROVED 123 LIBRARIES CAN'T CRASH — AND FOUND HUNDREDS OF BUGS

Anthropic's Frontier Red Team, working with the verification startup Theorem, says Claude wrote machine-checked formal proofs that 239 fuzz-test entry points across 123 open-source libraries can never crash on any input under 64 MB — about 180 million lines of Rocq proof, checked by machine, targeting the compiled RISC-V binary rather than the source code.

The proof hunt found hundreds of bugs fuzzers had missed, including a decompressor flaw that only fires on a rare 20 MB input and a rediscovery of CVE-2026-56132 in libexpat; in one PDF reader Claude found as many security bugs in a week as are typically reported in a year. Anthropic is blunt about the limits: the proofs cover single-threaded static RISC-V binaries on Linux, not the x86 and ARM systems most people run.