- Anthropic’s Frontier Red Team, working with partner Theorem, used Claude to generate roughly 180 million lines of mathematical proof code verifying the correctness of 123 open-source libraries in just a few weeks, The Decoder reported October 23, 2026.
- While generating proofs, Claude caught hundreds of bugs that conventional testing missed, including a logical error in a widely used PDF reader that had gone undetected for over a decade.
- Anthropic calls the process “fully autonomous,” though the underlying formal-verification tools and foundations were built by humans at Theorem; an independent program called Rocq checks every proof automatically.
- The proofs only hold under strict conditions, such as a specific processor architecture and single-threaded programs with inputs under 64 MB, and individual proofs can cost hundreds of millions of tokens to generate.
What Happened
Anthropic said Claude mathematically proved that existing software is correct by translating crash tests from 123 open-source libraries in Google’s OSS-Fuzz security project into formal mathematical problems, then generating proofs that an independent verification program called Rocq checked automatically, according to The Decoder.
Why It Matters
Formal verification has historically been reserved for extreme cases; a small operating system kernel called seL4 took a team of specialists 20 person-years to verify. If Claude can do comparable work in weeks, Anthropic argues the question shifts from whether AI can write code to whether it can prove that code is safe, a shift that could cut the time developers spend reviewing AI-generated software.
Technical Details
Verifying a widely used PDF reader turned up as many security bugs in one week as are typically reported for that software in a year. The proofs are also reusable: Anthropic says its newer Opus 5.5 model solves half of them with about 40 percent fewer tokens than Opus 5, echoing gains Claude Mythos showed finding 423 Firefox bug fixes in a month.
Who’s Affected
Maintainers of the 123 open-source libraries tested directly benefit, and Anthropic says the wider software industry could eventually treat formal verification as routine, similar to how AI tools recently surfaced decades-old PHP vulnerabilities missed for years.
What’s Next
Anthropic acknowledged “formally verified” does not mean error-free, since proofs only hold under specific hardware and input conditions, and it has not disclosed the project’s total compute cost, leaving open how affordable this approach will be at wider scale.
