Claude generated machine-checked security proofs across 123 open-source libraries
Anthropic and Theorem report that Claude produced formal security proofs at the RISC‑V machine-code level for 239 fuzz-test entry points across 123 open-source libraries.
AT A GLANCE
Key points
- Claude verified 239 entry points across 123 open-source libraries.
- The proofs cover the no-crash property for every input smaller than 64 MB.
- The project produced roughly 180 million lines of Rocq proofs.
- The findings are bounded by specific RISC‑V and Linux model assumptions.
What is changing?
Anthropic and Theorem published a large-scale formal software-verification effort on October 22. The team converted existing fuzz tests from open-source projects into mathematical properties, then had Claude write Rocq proofs showing those properties always hold in compiled RISC‑V code.
After patching bugs it found, Claude proved that 239 entry points across 123 OSS-Fuzz libraries do not crash for any input smaller than 64 MB. The generated proofs total roughly 180 million lines of Rocq; a typical proof contains about 600,000 lines and is checked by the external Rocq verifier.
The result does not mean the software is free of every bug. The proofs depend on assumptions including single-threaded static RISC‑V binaries, a fixed memory budget and modeled Linux behavior. Anthropic is releasing a subset of the proofs and trusted components for independent verification.
Explore the original sources
Source date: 22 October 2026.
Published by AI Radar: 23 Oct 2026, 06:59 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…