Image: WIRED
Why it mattersA team releasing AI-written code can soon buy a tool that proves parts of it correct. The honest question is which parts, because no vendor claims to verify a whole product.
WIRED reports that &Truth launched on Thursday with $70 million raised. The company wants to use AI to make formal verification work on any programming language. Its chief science officer is Ilya Sergey.
Formal verification is math that proves a piece of software does what its specification says, WIRED explains. It is slow and expensive by hand. The pitch is that generative AI makes it practical.
The founder and the technical hook
Chuyue Sun, a Stanford researcher, joined &Truth in August, WIRED reports. Her earlier work used AI to verify part of the code that encrypts conversations in the Signal messaging app. That demonstration is cited as the proof that AI-assisted verification can reach useful code.
"Bridges, buildings, they don't collapse because people did math before constructing them," Sergey told WIRED. "Why don't we do the same for code? The technology to mathematically reason was all there, just difficult to scale."
The company's stated aim is to eventually verify any programming language. WIRED does not name &Truth's investors in the extract read for this item.
The field is crowded already
Rival Harmonic has raised $295 million, WIRED reports. Robinhood chief executive Vlad Tenev co-founded Harmonic. Its math-focused AI agent, Aristotle, supports verifying C code and hardware designs written in Verilog. The agent is being tested by chipmakers Tensordyne and Nvidia, WIRED says. Nvidia's venture arm is one of Harmonic's investors.
Anthropic and the verification startup Theorem published a joint result last week. They used AI to help find bugs in open source software, WIRED notes. Trillions of ordinary tests had failed to detect those bugs. The two also launched an add-on for Claude for other developers. Costs were not disclosed, the article says.
What the sceptics in the same article say
Will Wilson, who runs the bug-hunting startup Antithesis, calls formal verification "a trap" for math nerds, WIRED reports. Antithesis spent millions of dollars verifying some of its own cybersecurity defences. Wilson describes the work as "a pain in the ass".
The piece ends on an open question. The quoted executives agree that AI makes formal verification cheaper. They disagree on whether it reaches the whole product, or only the few features that need the strongest assurance.
What this changes for a team shipping AI-written code
A team pushing AI-generated code into production can soon buy a tool that proves properties of its most sensitive parts. The real decision is which parts. No vendor cited in WIRED claims to verify a whole application. Pick the small set where a bug would be the worst. Write the specification in plain English. Then test what the verifier gives you back against your current test suite.
The second decision is where this work lives. &Truth sells it as a product. Anthropic sells it as a Claude add-on. In a year, the honest answer may be that the verifier runs inside the coding assistant.
Source
The 'Pain in the Ass' Math That Could Save Vibe Coding From Itself, WIRED.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
