ChainCatcher report: On October 22, Anthropic’s Frontier Red Team released a study stating that large language models can now automatically generate formal proofs at a larger scale. In collaboration with Theorem, the team used Claude to formally verify safety properties for over 100 real open-source libraries at the RISC-V assembly level, marking the first known instance of AI performing large-scale formal verification on production software at the assembly level. During the proof generation process, previously undetected vulnerabilities spanning multiple years were discovered. The experiment converted existing compilable fuzz tests from Google’s OSS-Fuzz into formal specifications: under assumptions of single-threaded, statically linked binaries, fixed memory budgets, RISC-V CPUs, and Linux environments, the program must exit cleanly and not trigger any memory safety violations or assertion failures for any input under 64 MB. Within weeks, after patching identified issues, Claude formally proved these properties for 239 entry points across 123 OSS-Fuzz libraries. The proofs were verified against compiled RISC-V instructions—approximately 180 million lines of Rocq—checked by the Rocq kernel. The page notes that this does not imply the libraries are entirely free of vulnerabilities; remaining issues may reside in trusted components or outside the scope of the proofs. Additionally, the team had Claude formalize RFC 1950 and prove that the zlib implementation in the Linux kernel conforms to this specification at the x86 binary level; they also developed an interpreter written in Rust that covers a large subset of Python, achieving 99.95% behavioral consistency with CPython across over 1.7 million randomly generated programs. On 36 small verification tasks, Opus 5.5 required approximately 40% fewer tokens than Opus 5 to prove half of them. The team also noted that the proofs apply only to single-threaded, statically linked RISC-V binaries, and the proof checker cannot determine whether the specifications themselves reflect real-world requirements.
Anthropic Uses Claude to Conduct Large-Scale Formal Verification on Open Source Libraries
ChaincatcherShare
Summary
On-chain news broke on October 22 as Anthropic’s Frontier Red Team revealed that large language models can now automatically generate formal proofs at scale. Working with Theorem, they used Claude to verify over 100 open-source libraries at the RISC-V assembly level. The team converted Google OSS-Fuzz fuzz tests into specifications and proved 239 entry points across 123 libraries. Proofs targeted RISC-V instructions and were verified via the Rocq kernel. Anthropic also formalized RFC 1950 and proved that Linux’s zlib conforms to it at the x86 binary level. A Rust interpreter achieved 99.95% consistency with CPython on 1.7 million programs. AI and crypto news continues to show breakthroughs in formal verification.
Source:Show original
Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information.
Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.