
Whitepaper · November 2026
Agents Under Control: Agentic AI in Public Workflows, With Audit Trails and Stop Controls
Who is allowed to act? What governments and regulated enterprises should require before an AI agent touches a public workflow, with a four-page toolkit.
- 26 pages
- Five chapters
- Four-page toolkit
Put AI agents under control in five chapters
- From answers to actionsWhat changes when AI stops answering and starts doing work.
- The risks, sourcedPublished risk lists, attack tests and what operators report.
- The controls to requireIdentity, narrow rights, approvals, logs, limits and tests.
- Decide what an agent may doAutonomy levels per workflow, and what governments have published.
- The toolkitA release scorecard, an agent register, an autonomy worksheet and a stop drill.