Anthropic’s Claude Automates Complex Code Verification

Anthropic has demonstrated that its Claude model can now mathematically verify the safety of existing software with minimal human intervention. The company worked with partner Theorem to create a system that generates formal proofs of code correctness. This marks a significant step in automating one of the most difficult aspects of software development.

The core claim is that AI can handle the rigorous task of formal verification, which has traditionally been too expensive for widespread use. In the past, proving that software runs error-free required armies of specialists. For example, a team needed 20 person-years to verify the seL4 operating system kernel. That cost limited the method to critical systems like aviation software. Anthropic and Theorem wanted to change that equation.

The team used Claude to analyze 123 open-source libraries from Google’s OSS-Fuzz security project. In just a few weeks, the model produced roughly 180 million lines of proof code. It did not write new software. Instead, it translated crash tests into mathematical problems and analyzed compiled machine code. An independent program called Rocq automatically checks these proofs. Anthropic describes the process as fully autonomous, though human experts at Theorem built the underlying tools.

A major benefit emerged during the testing phase. Claude discovered hundreds of bugs that conventional testing had missed. These flaws often only appear under specific conditions, such as unusually large input files. In one instance, the team found as many security bugs in a widely used PDF reader in one week as are typically reported in a full year. In another case, Claude caught a logical error that had remained undetected for over a decade. Even when a complete proof fails, the attempt often surfaces real issues.

However, there are important caveats. Anthropic notes that formally verified does not mean error-free. The proofs only hold under strict conditions. They require a specific processor architecture, single-threaded programs, and inputs smaller than 64 MB. In real-world scenarios, these libraries often run on different hardware and configurations. This gap between theoretical proof and practical application remains a key challenge.

The computational cost is also significant. Individual proofs can consume hundreds of millions of tokens. Anthropic has not released the total project costs. Despite this, the approach offers potential savings in the broader development cycle. If developers can trust AI-generated software more, they spend less time on manual review. A smaller test run showed that proofs are reusable. Newer models are also more efficient. Opus 5.5 solves half the proofs with about 40 percent fewer tokens than Opus 5.

The project shifts the focus from whether AI can program to whether it can prove its work is safe. Anthropic expects formal verification to become as routine as standard testing. Theorem’s roadmap already targets critical infrastructure components like glibc, OpenSSL, and cURL. If successful, this could transform how developers build and trust software systems.