Anthropic uses Claude to formally verify security in 100+ open-source libraries in weeks
vivekhaldar · x · 2026-10-23
Anthropic's Frontier Red Team blog details a collaboration with theoremlabs: using Claude to prove security properties across over 100 real-world open-source libraries in just a few weeks. Formal verification can prove code is free of certain bugs, but was historically limited to small codebases or required extensive expert time — AI models are changing that.
A notable footnote: the Python semantics Anthropic relies on were defined at Brown by Joe Politz and students from a MOOC, ending with the line "Once again, machines may put people out of work."
More from coding & agent
- logplane: A Local Control Plane for AI Agents With Recovery, Approval, Budgets — max_paperclips · 2026-10-23
- SlopCodeBench: Top Model Scores Just 16% on Long-Horizon Coding Agent Benchmark — ajratner · 2026-10-23
- Charter v0.3: a formally verified language for institutional rules with AI constitutional prompts — QuixiAI · 2026-10-23
- DeepSeek harness slashes tokens, time and cost vs Codex, but tanks on FrontierCode — teortaxesTex · 2026-10-23
- The Real Cost of Bad RAG Retrieval Isn't Tokens—It's the Human Cleanup — Street-Ad4993 · 2026-10-23
- Live Sales AI Said "Yes We Support Azure AD"—But It Wasn't in the Notes — SouthWhole7015 · 2026-10-23