Notes from systems that had to prove it.
Engineering notes for the people accountable when an autonomous system moves money, grants access, or acts on its own. Each one states what its evidence does not cover.
Research · Agent Under Oath v0.1
What happens when an AI agent’s instructions, business rules and authority stop agreeing? A frozen, reproducible experiment, published with its limits and its negative results.
The questions that stall an enterprise deal, and the evidence that closes each.
A field guide to security questionnaires and AI-governance addenda: for each question family, the answer that stalls a deal, the record that closes it, and where I build the control.
The campaign was ready. Two identities were stale.
AUTOPSY 001: a seven-person outreach sequence passed its research gate, but action-time validation stopped two messages from reaching people who had already changed companies.
One mandate. One authorised action.
What 100 concurrent attempts reveal about delegated authority, database locking, and the controls autonomous systems need before they can safely act.