CCDV-F

Question

A teammate is reviewing the team’s threat model for a Claude application and has asked you to identify the AI-specific threat categories the model should cover. The teammate has already listed traditional web-application threats and wants to know which additional categories apply to a Claude application.

Which AI-specific threat categories would you add?

Explanation

Claude applications have LLM-specific risks: prompt injection can redirect model behavior, prompts and context can disclose sensitive data, jailbreak attempts can circumvent guardrails, and unsafe generated output can bypass or undermine application controls. Anthropic identifies jailbreaks and both direct and indirect prompt injection as distinct threat models, and recommends limiting the model’s access to sensitive data and actions.

Learn more

Mitigate jailbreaks and prompt injections

CCDV-F

AstroTutor

Notes

Readiness

QuestionQ16

Community Discussion