&Truth launches to automate formal verification for software

In short: A new startup called &Truth has launched, betting that AI can make formal verification, a math-based way to check software, easier to use.

What happened

&Truth publicly launched this week, with Stanford researcher Chuyue Sun joining the company in August. The company says it wants to use generative AI to speed up “formal verification,” a process that uses math to prove software behaves the way it is supposed to.

Formal verification is often described like doing a careful proof in a math class. Engineers must write very precise statements about what the software should do, then check the code against those statements. It can find problems that normal testing misses, but it is slow and expensive.

Sun reported early results in a recent paper. She and other researchers used an AI tool to complete an engineering project in less than half a day for under $500, work that previously took three experts five months to do by hand. Sun also used AI to help verify part of the encryption code used by the private messaging app Signal.

&Truth is not alone. Rival startup Harmonic said this week that its math-focused AI agent, Aristotle, can support verification for software written in C and for hardware designs written in Verilog. Anthropic and the verification startup Theorem also said they used AI to find bugs in open source software that “trillions of tests” had not caught.

Why it matters

More software now runs critical systems, from phones to cloud services, and bugs can lead to security breaches. Supporters argue that math-based checks could make software more reliable, like inspecting a bridge design before building it. Critics say it can still miss issues caused by user mistakes or hardware failures, and it is unclear how widely companies will pay to use it.

Source: Wired