Anthropic uses Claude to formally verify 123 open-source libraries

Anthropic reports that Claude was used to produce formal proofs of security properties at the RISC-V assembly level for over 100 real-world open-source libraries. The work converts existing fuzz tests into specifications stating that the entry point never crashes for inputs under 64 MB, under stated assumptions about the runtime environment. Claude also found and patched bugs while writing the proofs. The proofs are checked by the Rocq kernel and are being published alongside the RISC-V and OS semantics.

Key takeaways

• Claude proved no-crash properties for 239 fuzz entry points across 123 OSS-Fuzz libraries, for inputs under 64 MB.
• Proofs were conditioned on assumptions such as single-threaded static binaries on RISC-V with Linux and a fixed memory budget.
• While writing the proofs, Claude discovered bugs that had gone undetected for years despite extensive testing.
• Existing fuzz tests were automatically converted into formal specifications over compiled RISC-V instructions.

Why it matters

If LLMs can produce machine-checked proofs at scale, formal verification may become practical for production software rather than only mission-critical systems, offering stronger guarantees than fuzzing alone.

Read with care

The guarantees hold only under the stated assumptions and input bound, and rely on trusted components such as the RISC-V and OS semantics being correct; source claims not independently checked.