Claude proves security properties in 123 open-source libraries

Claude, Anthropic's large language model, formally proved security properties across compiled code in 123 open-source libraries. The effort uncovered bugs that had gone undetected for years, demonstrating that LLMs can tackle formal verification tasks at a scale and speed previously impractical with manual methods.

Formal verification—proving that code meets security and correctness specifications—has long been a labor-intensive process limited to critical systems. By automating parts of this work, Claude shows AI can reduce the gap between what gets verified in production software and what security experts would ideally verify if resources were unlimited.