Public-benefit startup &Truth said Thursday it has emerged from stealth with $70 million in seed funding to use artificial intelligence for formal software verification, according to Wired and a company release carried by Singapore's Economic Development Board^2,3.

Formal verification turns programs into mathematical statements that can be checked against a specification, a process companies have long used mainly on the most critical code². &Truth said it wants a universal verification machine that can read code in any language and produce proofs others can check independently³.

Stanford computer scientist Chuyue Sun joined the company in August, Wired reported². A recent paper she co-authored said an AI tool finished a verification task on encryption code used by the Signal messaging app in less than half a day for under $500, compared with five months of work by three experts².

The company release, dated Oct. 29, said &Truth was founded in early 2026 and is led in part by co-founder and co-CEO Clark Barrett³. Ilya Sergey is chief science officer, Wired said². The release did not name the seed investors³.

Wired placed &Truth in a broader push: Harmonic, which has received $295 million and counts Nvidia's venture arm among investors, said this week its Aristotle agent can verify C software and Verilog hardware and is evaluating the tool with potential users including chipmakers Tensordyne and Nvidia². Last week Anthropic and Theorem said they used AI to help find bugs in open-source software that extensive testing had missed, Wired reported².

Whether the method will spread beyond organizations such as NASA and Apple remains open, Wired said². Demand is high, but pricing, how customers would adopt the tools, and which bugs proofs can actually rule out are unsettled². Will Wilson, who runs testing startup Antithesis, said his company still found errors after spending millions to formally verify some defenses, and found bugs in software customers had already verified².

The company release claimed AI can speed traditional proof work by 100 times or more and that &Truth's automation adds another 100 times, cutting generation from months or years to hours or minutes³. Those speedup figures were not independently verified in the coverage^2,3. Proofcraft executives told Wired that customer interest already outruns what existing verification firms can supply, while warning that AI-written specifications could create unjustified trust².