●builderYou should use these 56 attack scenarios to stress-test agent autonomy before deployment.
●researcherThis offers a more realistic assessment of IPI than single-step tool call evaluations.
●policyThis provides a structured taxonomy for assessing agentic security risks.