Reading Path4 steps
AI Authority in Cyber-Physical Systems
Focus on what an AI system is allowed to read, infer and change when its output can affect engineering or operational state.
Decision context
Use this path when AI can influence engineering or operational state. The decision is not whether the model is capable, but what it may read, infer and change, and which independent controls limit authority when inputs or dependencies are hostile.
01
Start here · AI Security · Product Security
AI Agents Turn Excessive Permission into Operational Risk
Define the authority problem before discussing agent capability.
→
02
Hostile input · AI Security · Product Security
Prompt Injection Is an Authority-Boundary Failure
Treat untrusted content as a control-path problem, not only a model-quality issue.
→
03
Dependency trust · AI Security
When AI Crossed the Trust Boundary
Extend the boundary into models, platforms and external artefacts you do not directly control.
→
04
Listen · Podcast
When AI Crossed the Trust Boundary: The OpenAI–Hugging Face Incident
Use the transitive-trust episode to revisit the governance decision.
→