{
  "id": 20544215,
  "title": "AI Is Starting to Help Prove Code Correct, Not Just Find Bugs.",
  "url": "https://urgent.news/2026/11/18/ai-is-starting-to-help-prove-code-correct-not-just-find-bugs",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-18T04:28:20.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/xabdullahahmad/ai-is-starting-to-help-prove-code-correct-not-just-find-bugs-407h"
  },
  "original_language": "en",
  "account": "Anthropic has recently collaborated with Theorem to utilize Claude, an AI model, for formally verifying security properties in 123 open-source libraries. This marks a shift from conventional testing methods, which only check software against specific inputs to uncover failures. Formal verification, on the other hand, employs mathematical proofs to establish the truth of certain properties across defined ranges of inputs. In this case, Claude generated proofs demonstrating that the tested programs would not crash when handling inputs under 64 MB, within a simulated Linux environment and under specific assumptions. Claude also identified and assisted in fixing numerous bugs, some of which would have been difficult to detect using traditional fuzz testing. Historically, such formal verification has demanded significant specialist effort, with the verification of the seL4 operating system kernel reportedly requiring approximately 20 person-years, compared to about two person-years for its implementation. Remarkably, Claude managed to complete the large-scale verification task within a matter of weeks. However, it is important to note that this process still required a vast number of tokens, presenting cost and efficiency challenges. Importantly, this does not imply that all 123 libraries are free of bugs. Instead, the proofs cover specific properties and code paths under particular conditions, rather than every conceivable flaw in the software. The potential impact of this development is significant, as AI could render mathematical verification more feasible for a broader range of software projects. This could allow developers to supplement testing efforts with AI-assisted formal verification, establishing more specific guarantees about the behavior of their code. For further details, refer to the research available at https://www.anthropic.com/research/formal-verification.",
  "summary": "Anthropic recently worked with Theorem to use Claude to formally verify specific security properties across 123 open-source libraries, covering 239 test entry points. The interesting part is how this differs from ordinary testing. Testing checks software against selected inputs to find failures. Formal verification uses mathematical proofs to establish that a specific property holds across a…",
  "key_points": [
    "Anthropic and Theorem use Claude to verify security properties in 123 open-source libraries.",
    "Claude generates proofs showing programs won't crash with inputs under 64 MB.",
    "Formal verification historically required 20 person-years, Claude completed in weeks."
  ],
  "editors_take": "Automating mathematical proofs with large language models lowers the barrier to formal verification, allowing developers to provide specific security guarantees that previously required years of specialized human effort.",
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}