&Truth Raises $70M to Let AI Prove Your Code Isn't Broken

&Truth stepped out of stealth on Thursday with $70 million in funding and a blunt pitch: use generative AI to mathematically prove code does what it claims. Stanford researcher Chuyue Sun, who joined the company in August, had already shown the approach works — her paper used AI to verify part of the encryption code that secures conversations in Signal. Chief science officer Ilya Sergey frames it simply: bridges and buildings don't fall down because engineers did the math first. Code, he argues, deserves the same treatment, and the only thing missing was scale.

The receipts come from Sun's own research. An AI tool finished an engineering project in less than half a day for under $500, according to her recent paper. Three highly skilled experts needed five months to do the same work by hand. That gap is what's dragging formal verification — long dismissed as too slow and too expensive — out of the academic shadows.

Competitors are moving fast. Harmonic, cofounded by Robinhood CEO Vlad Tenev, has raised $295 million and this week said its math-focused AI agent Aristotle can verify software written in C plus hardware designs in Verilog. The startup is testing the tech with potential users including chipmakers Tensordyne and Nvidia, whose venture arm is an investor. Just last week, Anthropic and verification startup Theorem said they used AI to surface bugs in open source software that trillions of tests had missed, and shipped a Claude add-on for developers doing verification work.

Not everyone is sold. Will Wilson, who runs bug-hunting startup Antithesis, calls formal verification a trap for math nerds like himself. His company spent millions verifying its own cybersecurity defenses — a process he describes as a pain in the ass — then still found extra errors with its own tool, and caught bugs in software that customers had already formally verified. Ethereum creator Vitalik Buterin has made a similar point, writing this year that formal verification is not a panacea, though he still sees it as a key defense against AI-driven cyberattacks.

The tech already quietly sits inside some of the world's biggest systems. Boeing planes, Amazon servers and more than 2.5 billion active Apple devices run at least some formally verified code, typically on the security side, according to company disclosures. Byron Cook, a vice president at Amazon Web Services, says verification has eased customer worries about cloud security and helped internal teams squash big bugs early. Apple's security team has said some corecrypto issues could never have been found any other way — while also admitting it has to stay conservative about adding new code so it can keep up with reviews.

Plenty of obstacles remain. Midsized tech companies tell WIRED they typically verify one key feature at most, citing thin verification teams, costs that outweigh the risks, and fears that waiting weeks for a proof will stall innovation. In open source, the nonprofit Beneficial AI Foundation used an Anthropic grant to validate the same encryption code Sun tackled, and maintainer Michael Rosenberg hopes sustainable funding appears — without leaning on AI companies' goodwill. Meanwhile Madan Musuvathi left Microsoft Research after 20-plus years to found Truidic, spurred partly by incidents like the OpenAI hack of Hugging Face. Proofcraft CEO June Andronick and chief scientist Gerwin Klein say inquiries now outpace what the industry can deliver, but warn AI-generated specifications could breed unjustified trust in the results.