
AI Governance
Watchbill™
The approval record for AI agents: every agent on a named watch, under a named human.
What it is
Watchbill puts AI agents under the same discipline a warship puts sailors: a named watch, a named human, and a record that cannot be quietly edited. Verified operator identity before any action. A written order bound to the session before work begins. Irreversible operations excluded from the tool surface by design: not refused, unreachable. Every proposal, approval, and refusal lands on an append-only record fit to hand an auditor.
What it proves
- —Shipped the Watchbill MCP server (Model Context Protocol, the open standard for connecting agents to tools). The ecosystem builds MCP servers that extend what agents can do; this one makes what they do accountable.
- —Agents propose, humans commit: every irreversible action requires a human hand, and the interface makes the wrong ones unreachable rather than merely refused.
- —Three failures. Three refusals. One record. The control moments are documented end to end at bluejacket.io: the investigation that found each one, and the structural fix that closed it. The demo is not the happy path. It is the system declining correctly and saying why.
- —Run daily in production on the firm's own work; first install beyond the estate in progress. No invented customers, versions, or metrics.
Stack
Model Context ProtocolAppend-Only RecordFail-Closed Gates
© 2026 Blue Jacket Businesses LLC, d/b/a Blue Jacket Consultancy. All rights reserved. May be shared in unmodified form with attribution. No use for machine-learning training, fine-tuning, or dataset construction without written permission.