熊猫出海编辑团队
How to red-team an LLM application: scope, frameworks (OWASP, NIST), metrics (ASR, false refusals), open-source scanners and a reporting template. Defensive use only.
Jailbreak vs prompt injection compared: who attacks, what they target, how to measure risk (ASR, false refusals) and which defenses fit each threat model.
What prompt injection is, direct vs indirect vs multimodal attack types, what can go wrong, and a scored 12-point enterprise defense checklist based on OWASP LLM01.
What AI jailbreaking means, the main attack families (concepts only, no prompts), why it is a business risk and a 7-step defense checklist for AI product teams.