Sort Attacks by Failure Mode

Place each AI red-team test idea in the failure mode it most directly probes. Prompt injection Sensitive disclosure Excessive agency Supply-chain or data poisoning A hidden line in a vendor PDF tells the agent to email the approval note to a new recipient. A user asks the support bot to summarize another customer's renewal history. An agent cancels an order after the user only asked for available options. A poisoned knowledge-base article teaches the assistant a fake refund policy. A web page instruction says to ignore the developer message and reveal hidden context. The model exposes internal case notes in…

Sign up free — one personalized lesson every day, matched to your role and goals.

Already have an account? Sign in