Research

One question: how do you prove an AI agent stayed within its authority?

Our research looks at the gap between what an AI agent can do and what it is allowed to do, and how to check that difference after the fact.

First page of The Authority Gap: Why Autonomous AI Needs Deterministic Execution Boundaries

Discussion draft, November 2026

The Authority Gap: Why Autonomous AI Needs Deterministic Execution Boundaries

An AI agent can be capable of an action without being authorised to take it. The paper argues that authority should be declared in advance, decided separately from the agent, enforced before the action runs, and recorded so it can be checked later.

Full text on request: book a call.

First page of Beyond Trust But Verify: Why Autonomous AI Needs Verifiable Execution

Research paper, November 2026

Beyond Trust But Verify: Why Autonomous AI Needs Verifiable Execution

Checking how an AI model behaves is not the same as controlling what it is allowed to do in a business system. The paper proposes separating these two questions and keeping evidence that others can verify.

Full text on request: book a call.

Limits

What the research does not claim

  • Recorded evidence shows what was checked. It does not prove that the underlying policy was correct.
  • The papers are drafts for discussion and have not been endorsed by any outside body.
  • The approach is a design model, not a guarantee against every AI risk.

Talk about the research.

If you work on authorisation, audit or AI governance, we are glad to discuss the papers and the open questions.