Claude Just Proved 123 Code Libraries Can't Crash

AI 📅 Oct 23, 2026 ✍️ Aamir Raza

Claude Just Proved 123 Code Libraries Can't Crash

Anthropic and Theorem used Claude to write machine-checked proofs that 239 entry points in 123 open-source libraries can't crash — and found hundreds of hidden bugs along the way. LexxWire

From Testing to Proof

Instead of checking inputs one by one, Claude wrote formal proofs a machine verifies — a guarantee, not just a test pass. LexxWire

180 Million Lines of Proof

The run produced about 180 million lines of Rocq proof, with a typical proof near 600,000 lines, per Anthropic's October 22 report. LexxWire

But Limits Apply

The guarantees cover compiled RISC-V programs under defined conditions — not proof the libraries are bug-free on every machine. LexxWire

Claude Just Proved 123 Code Libraries Can't Crash