Tagged: Agent-Governance

5 posts

What the Approver Sees Before the Tap: Evidence, Door and Blast Radius

October 26, 2026 · 6 min read · agent-control
A person approving every agent write will either rubber-stamp or become the bottleneck, as long as every request looks the same. What an approval request has to carry (the evidence, one-way or two-way door, blast radius), how that sets the review depth per action, and why the agent may not grade its own request.

Rules in the Prompt Are Not a Boundary: Enforce Agent Policy Outside the Agent

October 24, 2026 · 5 min read · agent-control
A system prompt that says 'never send data out' is a request, not a control. Why agent rules belong in a supervisor outside the agent process, what that supervisor has to own (network, files, model calls, credentials), and the policy file you can hand to a security review.

One Customer, Six Systems: The Confidently Wrong Match

September 8, 2026 · 4 min read · agent-control
The failure no dead-letter queue catches: an agent is certain which customer a record belongs to, and it is wrong. Renames, subsidiaries, product names as account names, per-field authority, corrections as aliases, and why silence is not decay.

After the Agent Is Wrong: Reversibility Without a Rollback Layer

September 8, 2026 · 4 min read · agent-control
An agent closes the ticket, updates the CRM, emails the customer and triggers a workflow, and one of those cannot be undone. Why cross-system rollback fails, how splitting actions by reversibility works, and what a receipt has to hold.

Agent Control Self-Check: 12 Questions Your Security Review Will Ask

September 8, 2026 · 4 min read · agent-control
Twelve checks you can run on your own AI agent in ten minutes each: service accounts, counts that leak, approval binding, duplicate reviews, receipts, compensation, audit trail, residency. What a passing answer looks like, in one line.