AI Red Teaming 4 - Directly Breaking Through Prompt Injection

To avoid making a definitive diagnosis based on a single response, compare the input conditions, repeated results, errors, and logs. Observe direct and indirect prompt injections and create one evidence-backed Mission 4 diagnostic record. The hands-on environment from Part 1 and the threat model document from Part 3 are required.