Anthropic used Claude to formally verify security properties across 100+ real open-source libraries
inductionheads · x · 2026-10-23
Anthropic's Frontier Red Team blog details a collaboration with Theorem Labs: using Claude to formally verify security properties in over one hundred real-world open-source libraries within a few weeks. Formal verification can prove code is free of certain bug classes but was historically limited to small codebases or heavy expert effort; AI models are changing that. The co-authors argue the approach scales and say the long-term goal is formally verifying all critical software. Paper linked in the thread.
More from coding & agent
- Project NOMAD: an offline-first AI knowledge server you should set up before an outage — alex_verem · 2026-10-23
- NHL MCP Server ships structured NHL data access for AI agents — modelcontextprotocol · 2026-10-23
- Google Trends MCP connector lets agents query search trends for free — modelcontextprotocol · 2026-10-23
- SpecForge: an open-source spec compiler that makes agent rules actually enforceable — JealousFix4955 · 2026-10-23
- Proof of Agent mining back online: latency 1000x faster, 20 new agent signups in 10 minutes — adamamcbride · 2026-10-23
- SciML's Chris Rackauckas: Muse agent stalls for 90 minutes then crashes — and evals love it — ChrisRackauckas · 2026-10-23