Over the last decade, AI has surged from speculative research to high-stakes deployment. During an AI Collective talk, Nexus business strategy manager, Alec James, highlighted a key imbalance: While AI capabilities grow exponentially, ethics and oversight lag far behind.
But here's a stark figure from his talk: In 2015, there were nearly 1,900 times more AI papers published than papers on AI ethics.
The result? An environment where powerful models operate with minimal transparency, accountability, or democratic input. As Alec noted, “Twenty to thirty people based in the U.S. cannot determine the alignment of such a powerful technology for the entire world.”
The case for verifiable AI
Verifiability, specifically cryptographic verifiability, is the foundation of the work happening at Nexus. Alec laid out its importance in stark terms: “You’re trusting the front-end, the team, and the prompt. But you have no idea what model or data you’re really interacting with.”
One solution is the Nexus zero-knowledge virtual machine (zkVM) that allows users to verify the integrity, correctness, and provenance of any AI computation — without re-running it and without revealing private data.
As Alec described, verifiability has immediate applications in:
AI-generated scientific literature: Preventing tampering and peer-review manipulation
Healthcare: Ensuring diagnostic models act with integrity and transparency
Credit scoring and insurance: Eliminating bias and opaque decision-making
Media provenance: Authenticating the origin of information in a world of deepfakes and misinformation
“With cryptographic verifiability, you don’t just check the output—you check the entire ingredient list that created it.”
Nexus Verifiable AI LAb
The Nexus Verifiable AI Lab is dedicated to exploring the frontier of verifiability, economics, and artificial intelligence in order to expand the boundaries of human and machine cooperation.
Alec traced the roots of verifiability to Alan Turing’s 1936 work on computability, positioning it as a parallel but neglected twin to the evolution of intelligent computation. While AI seeks to solve problems, verifiability proves whether those solutions can be trusted.
It’s a technical challenge — but also a philosophical one. In Alec’s words,
“Verifiability is about programmatic guardrails. You can gate AI outputs that don’t carry proofs of integrity from being used anywhere. That’s the real promise.”
The role Nexus plays
As part of its mission to build a verifiable intelligent internet, Nexus is developing a full-stack system to support trustless, privacy-preserving, globally verifiable AI:
A zkVM to prove the correctness of any computation
A distributed compute network to make proof generation efficient and accessible
A global blockchain layer to ensure immutable records and transparency
Nexus is already collaborating with leaders like AI Seer (Time’s Best Invention of 2024), Public AI, and Giza to bring verifiable AI to real-world systems—from fact-checking to autonomous agents .
As trust in AI systems feels fraught, and regulatory approaches struggle to keep up, verifiability offers a principled, decentralized path forward. It protects user privacy, ensures model accountability, and aligns incentives around integrity rather than opacity.
“We’ve seen what happens when ethics come second to profit. With AI, we don’t get a second chance. We need systems that are accountable by design.”
