Why Trustworthy Software Must Be Independently Verifiable: Chris Hsu

Based in Palo Alto and Singapore, &Truth is a public benefit corporation building on open-source formal verification research at Stanford University and the National University of Singapore. The PBC is building a universal verification machine to produce independently checkable mathematical proofs that software meets precisely defined safety, security and correctness requirements.

Founded in early 2026, the company is co-led by Christopher Hsu, Co-Founder, Co-CEO and Executive Chairman, who argues that no one should have to trust a company’s word about the validity of its proofs. As &Truth’s mission statement puts it, a verification layer the world can trust “can’t depend on any single company, including ours.”

If mathematical proofs become foundational to software security, banks, hospitals and governments could come to rely on them. But if only the company producing those proofs can validate them, that company becomes a new source of dependence. A technology designed to establish trust could instead concentrate it. &Truth’s answer is to make proofs transparent and independently checkable, so their validity does not depend on the authority or continued existence of any single provider.

 

Why Mathematical Proof Goes Beyond Expert Judgment

 

An audit or safety certification depends on expert judgment, documented evidence and defined standards. To trust either one, you must have confidence in the competence, independence and integrity of those conducting it. The strength of that assurance therefore depends on the scope, methods and credibility of the experts involved.

A mathematical proof works differently. Anyone with the proof, its definitions, dependencies and a trustworthy compatible checker can independently check its validity, regardless of who generated it. For example, Lean, an open-source proof assistant invented by Leonardo de Moura, uses a small, trusted kernel to check proofs against precise rules of logic. The kernel is deliberately simple compared with the sophisticated systems, including AI, that generate proofs. It applies the same rigorous standards whether a proof comes from a mathematician or an AI system.

Mathematical proof does not eliminate the need for judgment.

Someone must determine whether the formal specification faithfully captures what people actually intend and whether its assumptions reflect the real system. The checking kernel itself must also be trustworthy. The critical distinction is that these assumptions and obligations can be made explicit and scrutinized rather than relying solely on a vendor’s assessment.

That independence requires access to the proof, its definitions and dependencies, and a trustworthy checking tool. Without those elements, independent verification becomes impossible. The basis of assurance should be the mathematical proof and its explicitly stated specifications and assumptions, not the authority of the company that produced it.

Building Independent Verification Into &Truth

 

&Truth is architected around the principle that mathematical assurance should be independently checkable. Its verification engine builds on Lean, a free, open-source proof assistant. Proofs can be checked using publicly available software, provided the necessary definitions, assumptions and dependencies are accessible. Per its mission statement, &Truth is also committed to open-sourcing its foundational libraries and works with the Lean Focused Research Organisation, the nonprofit advancing Lean’s development. Hsu serves on its Strategic Advisory Board.

That transparency extends to &Truth’s &Atlas initiative, which aims to map the vast landscape of open-source C software. The project distinguishes code that has been mapped or analyzed from code whose specified properties have been mathematically proven. For meaningful independent scrutiny, each assurance should identify the exact software version, verified properties and underlying assumptions. A proof for one version does not automatically establish correctness after the code changes.

The company’s structure reflects a similar commitment. &Truth is a public benefit corporation, legally required to balance shareholder interests with its stated public benefit and the interests of those materially affected by its conduct. This provides a governance framework for its mission, although legal commitments alone cannot guarantee technical independence or public trust.

&Truth’s technical foundations draw on years of open-source research. Co-Founder and Co-CEO Clark Barrett, a Stanford computer scientist and professor, helped establish CSLib, an effort to develop shared formal foundations for computer science, analogous to Mathlib’s role in mathematics.

Co-Founder and Chief Science Officer Ilya Sergey founded a research laboratory at the National University of Singapore that developed Veil and Velvet, open-source frameworks for formally verifying distributed protocols and systems software, respectively. Both use Lean to produce independently checkable mathematical proofs. In November 2026, the laboratory released Veil 2.0, introducing formally verified model checkers for safety and liveness properties of distributed protocols.

Chris Hsu brings a complementary background in technology strategy, investing and AI safety. He is President of the Infinitude Foundation, which has supported the Lean Focused Research Organization and Stanford’s Center for AI Safety. Previously, he served as Founding Partner and CEO of Kilometre Capital and Managing Director at Citadel Investment Group. These experiences informed his approach to building reliable, transparent systems that can be independently verified.

 

Making Independent Verification The Standard

 

The foundations of critical infrastructure are established early. Decisions about openness, standards and control can determine whether a technology becomes widely accessible or dependent on a few providers. Once institutions build around those choices, reversing them becomes increasingly difficult.

If mathematical verification becomes essential to software security, the question will not simply be who can generate proofs fastest. It will be who can inspect the specifications, scrutinize the assumptions and independently verify the results. &Truth’s ambition is to make that level of assurance practical across the world’s critical software.

As Hsu contends, trust in software should rest on mathematical evidence that anyone can independently check, not on the authority of any single company.