Anthropic uses Claude to formally verify 100+ open-source libraries at assembly level
AxSaucedo · x · 2026-11-03
Anthropic's Frontier Red Team, partnering with Theorem, published a report showing LLMs make formal software verification feasible at scale for the first time.
- Claude produced machine-checked security proofs for over 100 real-world open-source libraries down to the RISC-V assembly level — described as the first large-scale effort applying formal verification to production software at that level
- While writing proofs, Claude uncovered bugs that had evaded detection for years despite trillions of test runs
- Motivation: AI makes finding vulnerabilities easy (the Linux kernel recently disclosed 1,000+ flaws), so proven guarantees are now needed; formal verification previously required years of expert effort
- Authors expect formal verification to become an essential part of secure software development as model capabilities improve
More from Safety
- UK High Court judgment cites non-existent case R v J, possibly AI-generated — andrewaltair · 2026-11-03
- Maybe AI excels at Navier-Stokes because fluid flows have no GDPR, researcher quips — lescarr · 2026-11-03
- Alleged 14T Anthropic "numbat" leak debunked: weights regenerate from name-seeded random noise — djcroman · 2026-11-03
- zkSNARKs proposed for verifying AI compute use: pacing enforcement and tamper detection — teortaxesTex · 2026-11-03
- Survey: 95% of APAC Leaders Claim They Can Explain AI Decisions, Only Half Can — mikeflache · 2026-11-03
- UK Government Publishes 116-Page AI Adoption Survey 2026 on Business Uptake — mikeflache · 2026-11-03