- Follow AI security frameworks such as MITRE ATLAS, the OWASP Top 10 for LLMs, and the NIST AI Risk Management Framework - Protect from direct, indirect, and cross-channel prompt-injection attacks - Prevent advanced jailbreaking and safety-filter bypass techniques - Protect from poisoned training data and RAG retrieval indexes - Prevent model-inversion, extraction, and model-theft attacks - Prohibit exploitation of multimodal weaknesses across text, image, audio, and video - Monitor for API flaws, unsafe model files, containers, and AI supply-chain risks - Know offensive AI tools and AI-assisted security workflows - Test MCP-related risks, including token theft and confused-deputy scenarios - Red-team AI guardrails through hands-on attack scenarios