1Prompt InjectionPrompt injection is what happens when an AI system cannot reliably tell instructions from the data it was asked to read.AI threats 2Jailbreaking & Guardrail BypassHow people try to push AI systems past their safety boundaries, why it works, and how to design defenses that do not rely on hope.AI threats 3Insecure Output HandlingInsecure output handling happens when software treats generated text or data as safe instructions without checking where it will be used.AI threats 4Excessive AgencyExcessive agency is what happens when a system can act more widely, independently, or consequentially than the job actually requires.AI threats 5Sensitive Information DisclosureSensitive information disclosure happens when a system reveals data to someone who should not see it, often through ordinary features, logs, or errors.AI threats