Claude proves existing code works correctly in weeks

Anthropic deployed its Claude model to mathematically verify that existing software code behaves correctly, a task that historically required vast amounts of expert labor. The company presented the work as a method to increase trust in AI-generated software, while clearly defining the limits of what the proofs guarantee. Claude did not write new software. Instead, it analyzed compiled machine code and constructed extensive proofs for existing programs.

Formal verification has long been expensive and slow. Specialists required 20 person-years to verify seL4, a small operating system kernel. Because of this cost, the technique remained rare, mostly reserved for critical systems like Airbus avionics. To change this dynamic, Anthropic’s Frontier Red Team partnered with Theorem. They handed the proof work to Claude, which generated approximately 180 million lines of mathematical proof code in a matter of weeks. The model drew on test programs from 123 open-source libraries in Google’s OSS-Fuzz security project.

A separate verification program called Rocq automatically checks the generated proofs. Anthropic stated that the process requires almost no human reviewers. The researchers described the proof generation as fully autonomous, though they noted that humans at Theorem built the underlying tools and formal foundations. During the work, Claude encountered hundreds of bugs and was allowed to fix them. Many of these flaws would likely have escaped conventional testing because they appear only under narrow conditions, such as unusually large input files or specific access sequences.

In one specific test, the team attempted to verify a widely used PDF reader. In a single week, they found as many security bugs as are usually reported for that software in a full year. In another application, Claude identified a logical error that had gone unnoticed for more than a decade. Anthropic added that even when a complete proof fails, the attempt can still reveal real bugs.

The company cautioned that formal verification does not mean the software is error-free. The proofs hold only under defined conditions, including a particular processor architecture, single-threaded programs, and inputs smaller than 64 MB. Real-world deployments often run on different hardware and configurations. Cost remains a constraint, as individual proofs consume hundreds of millions of tokens. Anthropic has not published total project costs but argues that verifying AI-generated code could reduce developer review time.

A smaller test run showed that proofs can be reused and that efficiency is improving. Opus 5.5 solves half the proofs using about 40 percent fewer tokens than Opus 5. The project shifts the focus from whether AI can program to whether it can prove the code is safe. Anthropic expects this method to become as practical as standard testing. Theorem’s roadmap already targets critical infrastructure including glibc, OpenSSL, and cURL.