Figure 10. AI agent framework with four attack surfaces and corresponding defense governance. AI: Artificial intelligence; LLM: large language model.