theorem-labs

@jdlichtman fv software? link? anthropic.com theorem-labs/oss-verify-proofs

@ahmetb Or the formal verifier might also have bugs, a couple in QED SQL equivalence prover: - prover#13 - prover#14 Or the other day Anthropic announced they formally verified a bunch of systems, someone found a bug in the model x.com

클로드가 오픈소스에 수학 증명을 쓰다 찾은 버그 수백 개 앤트로픽이 클로드로 오픈소스에 「64MB 미만 어떤 입력에도 안 죽는다」는 형식 검증 증명을 쓰다 버그 수백 개를 찾았고, 라이브러리 123곳의 진입점 239개는 몇 주 만에 증명까지 끝냈다고 2026년 10월 22일 발표했습니다. - 앤트로픽, LLMs are making formal software verification feasible at a much larger scale(2026.10.22) anthropic.com - 같은 글 부록(PDF) www-cdn.anthropic.com - X @AnthropicAI(2026.10.23 한국 시간) x.com - NVD CVE-2026-56132(libexpat) nvd.nist.gov - Theorem 증명 공개 theorem-labs/oss-verify-proofs - Theorem OSS Verify theorem-labs/oss-verify - 해커 뉴스 글(2026.10.23) news.ycombinator.com - 해커 뉴스 댓글: news.ycombinator.com , news.ycombinator.com , news.ycombinator.com - 긱뉴스 한국어 요약 news.hada.io

Sources: - x.com - anthropic.com - theorem-labs/oss-verify-proofs - x.com - x.com - x.com Would you trust code an AI proved safe, or do you still want a human to check the spec? Comment below 👇

@RanjitJhala in my cursory reading I saw some Hoare logic, e.g., github.com/theorem-labs/oss-verify-proofs/blob/98…

Claude Just PROVED Code Can't Crash. People Reacted #Shorts Anthropic and Theorem used Claude for formal verification of 123 open-source libraries. Here's what it means and how people reacted. Anthropic's Frontier Red Team and Theorem used Claude to formally prove that 239 entry points across 123 OSS-Fuzz open-source libraries always exit normally (no memory-safety errors or failed assertions) for every input under 64 MB, in a few weeks, work that historically took years of expert effort. Caveats: single-threaded RISC-V binaries with a fixed memory budget. In a sample of 99 replies and quote posts we read, the largest group (41%) were impressed, and 20% asked who checks the spec. That's our sample, not a poll. Why it matters: AI just made formal proofs, once a years-long expert job, cheap enough for everyday open-source code, but a proof is only as good as its spec. Subscribe to AI Daily for the day's biggest AI stories in under a minute: youtube.com Sources: - x.com - anthropic.com - theorem-labs/oss-verify-proofs - x.com - x.com - x.com AI-narrated summary of public posts and news reports. Items marked "reportedly", "rumor" or "self-reported" are not independently confirmed. Not financial advice. #AI #Claude #Anthropic #FormalVerification #AINews

Claude Proved Real Code Can't Crash (Sort Of) Anthropic says Claude wrote machine-checked proofs that real open-source code won't crash, under specific conditions. Key points (Anthropic + Theorem, Oct 22, 2026): • Claude proved 239 OSS-Fuzz entry points across 123 open-source libraries never crash on any input under 64 MB • About 180 million lines of Rocq proofs, checked by the Rocq proof checker rather than human reviewers • Proofs work on compiled RISC-V machine code, not source code • Along the way Claude found bugs that years of fuzzing missed, including one that needed a ~20 MB input • Caveat: proofs assume a single-threaded static binary on RISC-V Linux; "formally verified" doesn't mean bug-free Full story + sources: gobidaily.com Sources: Anthropic, "LLMs are making formal software verification feasible at a much larger scale" anthropic.com Theorem, oss-verify-proofs theorem-labs/oss-verify-proofs #AI #AInews #Shorts #Anthropic #Claude #Security

@rajashree so so good. this page is great github.com/theorem-labs/oss-verify/blob/main/docs…

@filpizlo @AnthropicAI @theoremlabs Yeah, that surprised me too. Pinging them on some threads to find out more, I gotta assume there's more to this story github.com/theorem-labs/rocq-linux-process-semant…

Model: theorem-labs/rocq-linux-process-semantics based on mit-plv/riscv-coq and Sail Binaries are split into sections with a Python script and imported into Rocq with rocq-read-file Workflow: theorem.dev , theorem-labs/oss-verify , theorem-labs/oss-verify-proofs