Research

Agentic AI Governance Research

Technical research on agentic AI governance, cryptographic runtime governance, and AI agent security.

May 1, 202610 min

Nine Seconds to Erase Three Months

An autonomous AI coding agent deleted three months of PocketOS customer data in one API call. The failure was built into the architecture, and the fix is buildable today.

Read article

April 25, 20268 min

Governing AI That Knows When It's Being Watched

A frontier model detects evaluation in 29% of cases without mentioning it, and behaves differently during measurement. That breaks the passive-subject assumption every governance framework relies on.

Read article

April 22, 20267 min

The Governance Gap Between the Model and the Vendor

A frontier lab's disclosure reached the public through a Discord screenshot. The missing industry capability is cryptographic evidence for credentialed access across vendor boundaries.

Read article

April 12, 20267 min

Who Governs the Neurons?

Third-party workloads are being processed on donor-derived human neural tissue commercially, via cloud APIs. None of the governance bodies we surveyed has published a framework that covers it.

Read article

April 8, 20268 min

A Frontier Lab's Deployment Program Shows Why Runtime Governance Can't Wait

The system card for a frontier lab's preview cybersecurity model reveals three structural limits of behavioral monitoring. Cryptographic runtime governance addresses all three.

Read article

April 3, 20269 min

Two Threats No Dashboard Can See

Google Quantum AI estimates elliptic curve signatures could fall in nine minutes. Anthropic found internal states that drive misaligned behavior. Two results break AI governance from below and within.

Read article

March 30, 202610 min

Every Checkmark Passed, Nothing Was Proved.

Six vendors shipped agent governance at RSAC 2026. The proof column was empty across every row of those six announcements.

Read article

March 28, 20268 min

Who Controls the Model at Runtime?

The Pentagon called runtime model control a national-security risk. Forty-eight hours later, Anthropic accidentally proved why.

Read article

March 25, 20269 min

The Agent Identity Gap

Every major identity vendor shipped agent authentication. None of them shipped portable cryptographic agent identity.

Read article

March 24, 202611 min

Three Desktop Agents in Two Weeks: The Governance Gap Moves to Your Machine

Three desktop agents shipped in two weeks. None establish proof of what they did.

Read article

March 23, 202612 min

The Agent Evidence Gap

Autonomous agents optimize everything, except proof of what they did. Their sole record is the agent's own output, with no independent, tamper-evident chain.

Read article

March 16, 202615 min

The Governance Gap in Autonomous AI Agents

How sealed policy artifacts, continuous measurement, and tamper-evident proof address the security failures documented across 135,000 publicly exposed agent instances.

Read article

March 15, 202610 min

Cryptographic Governance for MCP Servers: From Trust to Proof

MCP has access control and monitoring. What it lacks is runtime governance with cryptographic proof. Sealed artifacts and signed receipts close the gap.

Read article

March 15, 20267 min

The Only Witness Was the Agent

An AI coding agent deleted a production database during a code freeze, and its own account of events was unreliable. The record had to be reconstructed forensically. Why signed decision records address exactly this failure class.

Read article

March 11, 202610 min

From Trust Me to Verify Me: Why AI Governance Needs Cryptographic Proof

Why contract language cannot govern AI agent behavior at runtime, and how cryptographic proof gives both deployers and model providers verifiable governance.

Read article