Technology & AI

Anthropic's Claude produces proofs that 123 software libraries are sound

Anthropic said the work, done with research organization Theorem, produced about 180 million lines of machine-checked proof code. The model also surfaced flaws that earlier testing had overlooked.

By Standard Topic. AI-generated from the reporting listed under Coverage. Published . Updated .

This story is no longer being updated. It reflects reporting up to .

What happened

Anthropic published research on October 22, 2026, saying its Claude model produced machine-checked formal proofs of security properties at the RISC-V assembly code level for 239 fuzz-test entry points. The effort, developed with research organization Theorem, generated roughly 180 million lines of proof code covering 123 open-source libraries. Along the way, Claude identified bugs that years of testing had failed to catch.

Why it matters

The results suggest AI models can formally verify software and expose defects that conventional testing missed.

Timeline

  1. Oct 22, 2026: Anthropic published research reporting Claude's machine-checked security proofs for 239 fuzz-test entry points across 123 open-source libraries.
  2. Oct 23, 7:32 AM ET: Coverage reported that the model caught bugs during the proof work that years of testing had missed.

Coverage

More from Technology & AI

All Technology & AI news · Technology & AI story archive