What happened
&Truth, a public benefit corporation founded in early 2026, emerged from stealth on October 29, 2026, with $70 million in seed funding from formative AI safety and venture capital leaders. It is building a universal verification machine on Lean and Mathlib, a library of more than two million lines of mathematical definitions and theorems.
Why it matters
The company says its verification machine is designed to read code in any programming language and mathematically prove it is safe, secure, and correct, with proofs anyone can independently check—an approach that could shift software trust from testing or vendor claims to independently checkable proof.
WHO IT HITSEnterprise software development and security teams that rely on testing or vendor assurances could eventually use independently checkable proofs to verify AI-generated code, if &Truth's tools become available as planned.
Summaries like this, every morning.
&Truth is building on decades of formal-verification research at Stanford University and the National University of Singapore, and its launch materials tie the timing to a surge in AI-generated code. A companion essay cites Google CEO Sundar Pichai saying 75% of new code at Google is now AI-generated and approved by engineers, and Veracode's 2026 testing finding that AI models failed on nearly half of the coding tasks where security mattered. The same essay points to a September 2026 campaign in which an attacker deployed hundreds of AI agents to breach 395 organizations across 48 countries, and to a record 14,959 software vulnerabilities published that month—3.5 times the number a year earlier.
The company is positioning its approach against older formal-verification milestones that were too costly for routine use. The seL4 microkernel's original proof took roughly 20 person-years of expert effort for about 8,700 lines of code, and in 2015 under DARPA's HACMS program seL4 was built into Boeing's Unmanned Little Bird helicopter, whose flight controls a professional red team given weeks of access could not breach. More recently, in September 2026 an Anthropic AI model formalized the complete proof of Fermat's Last Theorem in Lean in 11 days, and earlier in 2026 Kim Morrison at the Lean FRO used Anthropic's Claude with minimal human guidance to build a Lean version of the zlib compression library and prove it always restores exactly the data it compressed.
&Truth says AI now accelerates formal proof work by 100x or more, and that its own proof automation adds another 100x, reducing proof generation from years or months to hours or minutes. Its 'specification factory' technology is meant to translate human intent and business requirements into precise formal specifications at scale. The company works with the Lean Focused Research Organization and intends to release its foundational libraries as open source, aiming for what it calls a shift from 'vibe-coding' to 'veri-coding,' in which AI-generated code is mathematically verified before deployment.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
