Claude generates 180 million lines of code to verify 123 software libraries

Anthropic’s Claude model generated approximately 180 million lines of mathematical proof code to verify the safety of 123 open-source software libraries. The project, conducted with partner Theorem, aims to prove that existing code behaves correctly without the massive human labor traditionally required for formal verification.

Formal verification has historically been too expensive for most developers. Specialists required 20 person-years to verify seL4, a small operating system kernel, which limited the technique to high-stakes environments like Airbus avionics. Anthropic’s Frontier Red Team changed this dynamic by assigning the proof generation to Claude. In a matter of weeks, the model analyzed compiled machine code and converted crash tests into mathematical problems.

The process relies on a separate verification program called Rocq to check the proofs automatically. Anthropic states that the system requires almost no human reviewers. While the researchers describe the generation as fully autonomous, they note that humans at Theorem built the underlying tools and formal foundations. Claude did not write new software but verified existing code by constructing extensive proofs.

The verification effort uncovered hundreds of bugs that Claude was permitted to fix. Many of these errors likely escaped conventional testing because they only appear under narrow conditions, such as unusually large input files or specific access sequences. In one instance, the team found as many security bugs in a widely used PDF reader during a single week as are typically reported in a full year. In another case, Claude identified a logical error that had remained unnoticed for more than a decade.

Anthropic emphasizes that formal verification does not mean error-free. The proofs hold only under specific conditions, including a particular processor architecture, single-threaded programs, and inputs smaller than 64 megabytes. Real-world deployments often run on different hardware and configurations, which may not be covered by these proofs.

Cost remains a factor, as individual proofs consume hundreds of millions of tokens. Anthropic has not published total project costs but argues that AI verification could reduce the time developers spend reviewing AI-generated software. A smaller test run showed that Opus 5.5 solves half the proofs using about 40 percent fewer tokens than Opus 5, indicating improving efficiency.

Theorem’s roadmap now targets critical infrastructure libraries, including glibc, OpenSSL, and cURL. Anthropic expects formal verification with AI to become as practical as standard testing, shifting the industry focus from whether AI can program to whether it can prove that code is safe.

Anthropic’s Frontier Red Team and Theorem completed the initial verification work for the 123 libraries from Google’s OSS-Fuzz security project. The project demonstrates that AI can generate the mathematical rigor needed to validate legacy software at a speed that was previously impossible for human specialists.