🛡️Stanford's AI Tool Verifies Software in Half a Day for Under $500

Formal verification just got cheap enough to matter

A Stanford researcher's AI tool just did in half a day, for under $500, what three experts needed five months to finish by hand. That's the kind of number that wakes up a field. The field is formal verification, the math-based method for proving software is bug-free, and a wave of startups now bets that generative AI will make it practical enough to use everywhere. If your code handles money, logins, or anything else where a bug costs real trouble, this one is yours. Formal verification means writing precise statements about what your software should do, then converting your code into formulas that can be checked. Historically that was brutal work. If AI handles the tedious parts, the process moves from a niche tool for high-stakes systems to something a normal team might run on a Tuesday. Anthropic and Theorem already used AI to find bugs in open source software that trillions of tests had missed. The startup money is real. &Truth launched publicly on Thursday with $70 million raised and wants to verify any programming language eventually. Rival Harmonic has $295 million and says its agent Aristotle now verifies C code and Verilog hardware designs. Harmonic is testing with chipmakers Tensordyne and Nvidia. Proofcraft, a long-running verification company, says inquiries are flooding in. The catch: formal verification proves code matches its spec, not that the spec is right. It also assumes things about how users and hardware behave, and some experts doubt it can catch user error or hardware failure. Watch whether these tools reach customers directly or through partners like Anthropic.