AdvertiseDonate
Latest20 Labour MPs Pay 9,600 Pounds for Leadership Training
ChicagoMilwaukeeMadisonFort WayneIndianapolisToledoDetroitCincinnati
News Locall · 330 North Wabash Avenue Suite 23rd Floor, Chicago, IL 60611
Tech · AI

180 million lines of proof code generated in weeks by AI

Anthropic and Theorem utilized Claude to mathematically verify open-source software, uncovering hundreds of latent bugs and reducing the need for manual review.

By Adrian Romano ·

Illustration for: 180 million lines of proof code generated in weeks by AI
Illustration generated for this article, not a news photograph.

Anthropic and its partner Theorem have generated approximately 180 million lines of mathematical proof code in a matter of weeks, marking a significant shift in how software correctness is established. This volume of formal verification work was applied to 123 open-source libraries, a task that traditionally requires years of specialized human labor.

The core objective of the project was to address the high cost and complexity of formal verification. Historically, proving that software operates without errors under all conditions has been prohibitively expensive. For context, a team of specialists required 20 person-years to formally verify the seL4 operating system kernel. Consequently, this rigorous method was largely reserved for critical systems such as Airbus avionics. By leveraging Claude, the Frontier Red Team aimed to automate this process, translating crash tests from the Google OSS-Fuzz security project into mathematical problems and analyzing compiled machine code to produce extensive proofs.

The system operated with a high degree of autonomy. Claude did not write new code but instead verified existing software. An independent program called Rocq automatically checked the generated proofs. According to Anthropic, the process required minimal human intervention, with researchers describing it as fully autonomous proof generation. However, the underlying tools and formal foundations were developed by humans at Theorem.

During the verification process, the model identified hundreds of bugs that conventional testing had missed. Many of these defects only manifested under specific conditions, such as unusually large input files or particular sequences of memory access. In one notable instance, the team found as many security vulnerabilities in a widely used PDF reader in a single week as are typically reported over an entire year. In another case, Claude detected a logical error that had remained undetected for more than a decade.

Despite the success, the approach has clear limitations. Anthropic notes that formal verification does not equate to complete error freedom. The proofs are valid only under strict conditions, including a specific processor architecture, single-threaded programs, and inputs smaller than 64 MB. Real-world applications often run on varied hardware and configurations, which may fall outside these parameters.

The financial implications of the project are also significant. The compute costs are steep, with individual proofs consuming up to hundreds of millions of tokens. While Anthropic did not disclose the total project cost, the efficiency gains suggest potential savings across the software development lifecycle. If developers can trust AI-generated code more readily, time spent on manual review decreases. This reduction in labor costs could make the entire coding process more economical.

The project also highlights trends in model efficiency. A smaller test run indicated that proofs are reusable and that newer models are becoming more cost-effective. Specifically, Opus 5.5 solved half of the proofs using approximately 40 percent fewer tokens than its predecessor, Opus 5. This efficiency improvement supports the argument that AI-driven verification is becoming a viable routine practice rather than a specialized tool.

The broader implication is a shift in the question surrounding AI programming. The focus is moving from whether AI can write code to whether it can prove that the code is safe. If the answer is yes, the impact could extend well beyond current coding tools. Anthropic expects formal verification with AI to become as standard as routine testing in the near future. Theorem’s roadmap already targets critical infrastructure components such as glibc, OpenSSL, and cURL, signaling a push toward integrating these methods into foundational software.

More Tech · Report a business · Got a Story?

Get News Locall in your inboxLocal news, all in one place. Free, a few times a week, unsubscribe any time.

Related stories

Get News Locall in your inboxLocal news, all in one place. Free, a few times a week, unsubscribe any time.