Brian Trunzo Urges ZK Proofs for AI Agents as 1,000 State Bills Trail the Threat
Updated
Updated · CoinDesk · Jul 17
Brian Trunzo Urges ZK Proofs for AI Agents as 1,000 State Bills Trail the Threat
3 articles · Updated · CoinDesk · Jul 17
Summary
zero-knowledge proofs should become the internet’s trust layer for AI, Brian Trunzo argues, giving every high-risk agent a cryptographic receipt for who authorized it, what data it used and what constraints governed it.
4% detector accuracy after simple image blurring shows why AI detection is failing, he says, while autonomous agents already buy, publish and negotiate in ways that can turn poisoned data into millions or billions in losses.
90-plus federal AI recommendations and more than 1,000 state bills introduced in 2025 are still geared to chatbots rather than agents, leaving regulation behind systems that can act before disclosures or labels matter.
HTTPS solved the 1990s web trust problem with cryptographic proof, and Trunzo says ZK can do the same for AI by verifying media provenance, model outputs and human-versus-agent identity without exposing private or proprietary data.
NIST is already exploring zero-knowledge standards through its privacy-enhancing cryptography work, which he says should underpin federal rules making lack of proof—not content itself—the basis for liability.
If AI can perfectly mimic reality, can we trust cryptographic 'proofs,' or are they the next technology to be compromised?
As China mandates AI agent traceability, is the U.S. falling behind on securing its digital economy from autonomous threats?
With AI agents creating a $26 billion liability crisis, who is legally responsible when autonomous code causes financial disaster?
The 2026 AI Trust Crisis: Over Half of Enterprises Impacted—How Zero-Knowledge Proofs Are Reshaping Security and Regulation
Overview
In 2026, the rapid spread of AI agents has led to a major trust crisis, with over half of enterprises experiencing security incidents or near-misses. This surge in vulnerabilities is driven by a 'deploy first, ask questions later' approach, where 69% of companies let agents share credentials and essential security measures lag behind. Most organizations invest little in dedicated AI security, relying instead on basic tools from model providers. As a result, AI agent deployment is outpacing proper safeguards, exposing businesses to growing risks and highlighting the urgent need for stronger, verifiable trust solutions.