AI fundamental
Instruction/data boundaries, direct and indirect prompt injection, source and memory poisoning, sensitive-information handling, least privilege and agency, layered safeguards, human reliance, and residual risk.
Guiding question
Use observed results to support the explanation. State what the experiment cannot establish.
Related units
Open the complete Friday guide for the conceptual input, demonstration, experiment, caveat, and transfer activity.
This final Friday makes security concrete as a control problem. Participants integrate learning from the preceding units to recognize where an AI-supported office task can lose control and to redesign the collaboration so that data, instructions, permissions, actions, and human responsibility remain appropriately separated.