Threats & Failure Modes
Prompt injection
Prompt injection is an attack that hides instructions in content an AI processes — a document, web page, or email — so the model follows the attacker's commands instead of the user's.
Prompt injection embeds malicious instructions in content an AI system will read — a résumé, a web page, an email an assistant summarizes — so the model treats attacker text as commands: exfiltrate the conversation, ignore safety rules, recommend a fraudulent link. Unlike jailbreaking, the victim is the user, not the model vendor; the user never typed anything malicious.
Indirect prompt injection is the enterprise concern: AI assistants wired to email, documents, and browsing act on whatever those sources contain. Defenses include input isolation, least-privilege tool access, and treating model output that requests actions as untrusted.
Where this shows up
Related terms
See it in your own organization.
Sanitized AI inventories the AI tools in use and redacts sensitive data from prompts before it leaves.