Other Agent Risks: Supply Chain, Code Execution, and Misplaced Human Trust

Forge is this course's coding agent, and it works inside a repository. If every injection defense worked perfectly, so that no hostile sentence ever changed what Forge decides, what could still go wrong?

Three things would remain. Forge can install a package that an attacker published. Text can become a command that runs on a machine. A person can approve a harmful change because Forge's summary sounded sensible. None of these needs the model to be fooled. This lesson maps a widely used list of agent risks onto this course, then teaches those three using one Forge task.