Palo Alto startup &Truth raises $70 million to mathematically prove software is safe
Led by a Stanford computer scientist, the company says AI can turn years of checking code into hours.
&Truth, a new company based in Palo Alto and Singapore, said Oct. 29 that it has raised $70 million to build a system that mathematically proves computer code works as intended. The company did not name its investors.
How much code, and how little is proven
GitHub, the Microsoft-owned site where most programmers store their code, said developers saved nearly 1 billion batches of code changes, called commits, there in the year ending August 2025, up 25% from the year before. That is about 2.7 million a day.
A growing share is written by AI. Google CEO Sundar Pichai said in April that 75% of all new code at Google is AI-generated and approved by engineers, up from 50% the previous fall.
Very little code is mathematically proven, and no one publishes a worldwide count. &Truth said hand-written proofs were so slow and costly that they were limited to "a handful of critical systems." One famous example, seL4, the 8,700-line core of an operating system, took about 20 person-years of proof work, its researchers reported. &Truth said AI and its own tools can cut such work from months or years to hours or minutes.
What "verified" means
Proving code correct, a field called formal verification, means using math to show that a program always does what a precise written description of its job, called a specification, says, for every possible input. Ordinary testing checks only the cases someone thought to try.
A proof rules out hidden backdoors or malware only if the specification forbids them. It also depends on assumptions, such as the hardware working correctly, the seL4 team wrote. Proof shows code meets "precise requirements across all possible inputs, under clearly stated assumptions," said Ilya Sergey, &Truth's co-founder and chief science officer.
What AI-written code means
GitHub said AI tools are letting more people build software, and Pichai said Google engineers now direct teams of AI agents instead of writing every line. But GitHub said the most common security flaw its scanner found in 2025 was missing checks on who may see or change data, up 172%, partly from AI-generated code. &Truth said AI code is being deployed faster than people can review it. "We must move from hoping code works to proving it does," said Chris Hsu, &Truth's co-founder and co-CEO.
Co-founder and co-CEO Clark Barrett is a Stanford computer science professor. &Truth is a public benefit corporation, a company legally required to pursue a social mission, and plans to share its basic tools for free.
Source: &Truth, "&Truth Emerges From Stealth," Oct. 29, 2026; GitHub, Octoverse 2025; Google, Sundar Pichai at Cloud Next 2026; seL4 team, "seL4: Formal Verification of an Operating-System Kernel".