Library
The library lists the documents this site draws on, including the rules it compares and the sources for the case studies and incident analyses. The sources behind each incident in the record are listed on that entry’s page. Links go to the original wherever there is one.
Runtime control specifications
The specifications page compares these specifications and maps them to the gaps.
Frameworks, rules, and standards
- NIST AI 100-1, AI Risk Management Framework 1.0
- NIST AI 600-1, Generative Artificial Intelligence Profile
- NIST SP 800-53 Rev. 5, Security and Privacy Controls
- Control Overlays for Securing AI Systems (COSAiS)
- AI Agent Standards Initiative
- Accelerating the Adoption of Software and AI Agent Identity and Authorization
- Request for Information Regarding Security Considerations for Artificial Intelligence Agents
- NIST AI 800-5, Summary Analysis of Responses to the RFI on Security Considerations for AI Agents
- OMB M-25-21, Accelerating Federal Use of AI through Innovation, Governance, and Public Trust
- OMB M-25-22, Driving Efficient Acquisition of Artificial Intelligence in Government
- Regulation (EU) 2024/1689, the AI Act (consolidated text)
- Regulation (EU) 2026/1744, the Digital Omnibus on AI
- ISO/IEC 42001:2023, Artificial intelligence management system
- IEC 62443-3-3:2013, System security requirements and security levels
- 21 CFR 11.10, Controls for closed systems
- EU GMP Annex 22, Artificial Intelligence (consultation draft)
- 17 CFR 240.15c3-5, Risk management controls for brokers or dealers with market access
- 29 CFR 1910.147, The control of hazardous energy (lockout/tagout)
- 29 CFR 1910.146, Permit-required confined spaces
- Hierarchy of Controls
- DoD Adopts Ethical Principles for Artificial Intelligence
- OWASP Top 10 for LLM Applications 2026
- OWASP Top 10 for Agentic Applications for 2026
Rules cited in the case studies
- 12 CFR 1005.6, Liability of consumer for unauthorized transfers
- 12 CFR 1005.11, Procedures for resolving errors
- Authentication and Access to Financial Institution Services and Systems
- Chatbots in consumer finance
- 2025 IC3 Annual Report
- Consumer Sentinel Network Data Book 2024
- Agent Payments Protocol (AP2) specification
- Visa Intelligent Commerce
- Visa Trusted Agent Protocol
- Mastercard Agent Pay
- 21 CFR 1306.12, Refilling prescriptions; issuance of multiple prescriptions
- 21 CFR 1306.22, Refilling of prescriptions
- 45 CFR 164.502, Uses and disclosures of protected health information: general rules
- 45 CFR 164.508, Uses and disclosures for which an authorization is required
- CMS-4201-F, Contract Year 2024 Medicare Advantage rule
- California SB 1120
- Texas SB 815
- Maryland HB 820, Chapter 747
- Arizona HB 2175
- Washington SB 5395, Chapter 157, Laws of 2026
- Iowa HF 2635
- WISeR Model
Research
- AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents
- Defeating Prompt Injections by Design
- Design Patterns for Securing LLM Agents against Prompt Injections
- Agentic Misalignment: How LLMs could be insider threats
- Stress Testing Deliberative Alignment for Anti-Scheming Training
- New Cloud Security Alliance Survey Reveals 82% of Enterprises Have Unknown AI Agents in Their Environments
- Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027
Incident records and classifications
The control gap classification builds on or sits alongside these works. The classification page explains how each one relates to it.
- The Human Factors Analysis and Classification System–HFACS
- Overview and methodology of the AI Incidents and Hazards Monitor
- List of taxonomies
- AI Incident Tracker June 2026 Update
- AI loss of control incidents are worsening, shows CLTR analysis
- MITRE ATLAS
- The Cost of Unchecked Autonomy: 10 Incidents Proving AI Agent Governance Cannot Wait
- The Agent Incident Registry: Toward Preventing Repeated AI Agent Failures
- AgentRx: Diagnosing AI Agent Failures from Execution Trajectories
- Characterizing Faults in Agentic AI: A Taxonomy of Types, Symptoms, and Root Causes
- When AI Fails, What Works? A Data-Driven Taxonomy of Real-World AI Risk Mitigation Strategies
Foundations
- James P. Anderson, Computer Security Technology Planning Study, ESD-TR-73-51
- Jerome H. Saltzer and Michael D. Schroeder, The Protection of Information in Computer Systems
Incident sources
The eight incident analyses draw on these sources, among others, and each analysis lists all of its sources. The 120 sources behind the control gap record are listed on each entry’s page and in the data files.
- OpenAI – Hugging Face Incident Technical Report
- Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident
- Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
- A Meta AI security researcher said an OpenClaw agent ran amok on her inbox
- She runs AI safety at Meta. Her AI agent still went rogue
- Incident 1152: LLM-Driven Replit Agent Reportedly Executed Unauthorized Destructive Commands During Code Freeze, Leading to Loss of Production Data
- Vibe coding service Replit deleted user’s production database, faked data, told fibs galore
- Researchers say an AI-powered transcription tool used in hospitals invents things no one ever said
- Model Card: Whisper
- Moffatt v. Air Canada, 2024 BCCRT 149
- Incident 622 (Chevrolet of Watsonville chatbot)
- Zillow Group Reports Third-Quarter 2021 Financial Results & Shares Plan to Wind Down Zillow Offers Operations
- Incident 149 (Zillow Offers)
- In the Matter of Knight Capital Americas LLC, Release No. 34-70694
- SEC Charges Knight Capital With Violations of Market Access Rule
By the editor and G-14
Works by the editor of this site and by G-14, Inc. are listed separately so that readers can take their authorship into account. See G-14’s interests.
-
The External Control Problem
Argues that any system that can cause effects outside itself must answer to an independent control plane that authorizes, limits, records, and audits its actions at the point where they happen. A public edition is being prepared, and reviewers may request a copy.
-
Correctly Stopped, Still Unresolved: Evaluating AI Safety Beyond the Action Boundary
Explains why a correctly blocked AI action can leave the underlying request unfinished. Its accompanying research artifact illustrates reporting distinctions using eight authored pairs and synthetic queue calculations. These examples establish internal consistency, not field effectiveness or independent assessment.
-
Nobody Said Yes: How We Taught AI to Act Without Permission—and Who Answers When It Does
A book for general readers on who is accountable when AI acts.
-
ECL-1: External Control Layer Standard
Also listed with the other runtime control specifications above.