- 工程边界与权限主轴：OpenAI
Structured Outputs/Safety in building agents、AnthropicDefine tools/Programmatic tool calling、GoogleStructured output/Function calling reference、MicrosoftStructured outputs/RBAC - 制度与监管边界主轴：
EU AI Act、Council of EuropeFramework Convention、NIST AI RMF、ISO/IEC 42001、OMB M-25-21、CanadaDirective on Automated Decision-Making - 安全与越权边界主轴：
Indirect Prompt Injection、InjecAgent、Security Challenges in AI Agent Deployment、Operator System Card、OpenAIDesigning AI agents to resist prompt injection、AnthropicClaude Code sandboxing、GoogleGemini Deep Research Agent - 规则、约束与结构校验主轴：
Constrained Language Models Yield Few-Shot Semantic Parsers、Grammar-Constrained Decoding for Structured NLP Tasks without Finetuning、ReLM、Logic-LM、Grammar-Constrained Decoding Makes Large Language Models Better Logical Parsers、Guardrails AI - 高风险行业边界主轴：WHO
AI ethics and governance guidance for large multi-modal models、FDAClinical Decision Support Software、Evaluation and mitigation of the limitations of large language models in clinical decision-making、ABAFormal Opinion 512、U.S. CourtsAnnual Report 2025、OCC Model Risk Management、CFPB Circular 2022-03 - 人机责任与审批链主轴：Anthropic
Building Effective Agents、MicrosoftApply generative orchestration capabilities/Transparency Note for Azure Agent Service/Task Adherence、GAO Accountability Framework、Human-AI Teaming、Reviewable Automated Decision-Making
附录 G：分章资料清单