The Chat Is Closed, the Agent Is Still Acting by Kai London
AI incidents do not look like traditional security incidents: a model gives harmful advice, an agent takes an action nobody intended, a system is quietly manipulated. This book sets out how to recognise, contain, investigate and report AI incidents before they become crises.
What the book covers
What counts as an AI incident
Failures, attacks, misuse and harmful outputs.
Detection
Signals, monitoring and reporting channels for AI problems.
Containment
Kill switches, rollback and limiting an agent's authority.
Investigation
Logs, prompts, data and model evidence.
Reporting and disclosure
Regulatory, contractual and public communication.
Learning and improvement
Post-incident review and strengthening AI controls.
This book offers general guidance and is not legal advice. It is independent and carries no endorsement by or affiliation with any regulator or standards body.
Who it is for
CISOs, incident response teams, AI engineering leads, risk managers and legal counsel.
About the author
Professor Kai London — CISSP, CISM.
An internationally recognised cybersecurity executive, board advisor and Founder & CEO of Quantum AI Systems Security LLC, writing at the convergence of AI, governance and operational resilience. Honorary Professor and Researcher at UCL.