mlsanityData to 7 Oct 2026

Practical challenges of control monitoring in frontier AI deployments

arXiv 2512.22154 ↗First posted 15 Dec 2025cs.CR cs.AI cs.MAcited by 6

Not on the map. Why?

Abstract (arXiv)

Automated control monitors could play an important role in overseeing highly capable AI agents that we do not fully trust. Prior work has explored control monitoring in simplified settings, but scaling monitoring to real-world deployments introduces additional dynamics: parallel agent instances, non-negligible oversight latency, incremental attacks between agent instances, and the difficulty of identifying scheming agents based on individual harmful actions. In this paper, we analyse design choices to address these challenges, focusing on three forms of monitoring with different latency-safety trade-offs: synchronous, semi-synchronous, and asynchronous monitoring. We introduce a high-level safety case sketch as a tool for understanding and comparing these monitoring protocols. Our analysis identifies three challenges -- oversight, latency, and recovery -- and explores them in four case studies of possible future AI deployments.

Key facts. Cited by 6 papers in the map (data through 7 Oct 2026). Builds on 22 of its 29 references that are in the map. First posted on arXiv 15 Dec 2025.

Builds on

22 of 29 references.

References · 22

  1. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,355
  2. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,330
  3. 2024MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineeringcited by 429
  4. 2025Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safetycited by 222
  5. 2023AI Control: Improving Safety Despite Intentional Subversioncited by 207
  6. 2024Large Language Model-Based Agents for Software Engineering: A Surveycited by 173
  7. 2025Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teamingcited by 172
  8. 2024LLM-Based Multi-Agent Systems for Software Engineering: Literature Review, Vision and the Road Aheadcited by 155
  9. 2025Auditing language models for hidden objectivescited by 68
  10. 2025International AI Safety Reportcited by 64
  11. 2024Safety Cases: How to Justify the Safety of Advanced AI Systemscited by 48
  12. 2025Ctrl-Z: Controlling AI Agents via Resamplingcited by 40
10 more
  1. 2025A sketch of an AI control safety casecited by 36
  2. 2024Towards evaluations-based safety cases for AI schemingcited by 32
  3. 2025Evaluating Frontier Models for Stealth and Situational Awarenesscited by 30
  4. 2024Games for AI Control: Models of Safety Evaluations of AI Deployment Protocolscited by 28
  5. 2024Adaptive Deployment of Untrusted LLMs Reduces Distributed Threatscited by 27
  6. 2025AI Agentic Programming: A Survey of Techniques, Challenges, and Opportunitiescited by 26
  7. 2025How to evaluate control measures for LLM agents? A trajectory from today to superintelligencecited by 20
  8. 2024Safety case template for frontier AI: A cyber inability argumentcited by 19
  9. 2024Subversion Strategy Eval: Can language models statelessly strategize to subvert control protocols?cited by 14
  10. 2025Combining Cost-Constrained Runtime Monitors for AI Safetycited by 10

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 6.

6 papers.

  1. 2026Training Agents to Self-Report Misbehaviorcited by 3
  2. 2026Distributed Attacks in Persistent-State AI Controlcited by 3
  3. 2026Classifier Context Rot: Monitor Performance Degrades with Context Lengthcited by 2
  4. 2026Retrying vs Resampling in AI Controlcited by 1
  5. 2026Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineagescited by 1
  6. 2026Red-Teaming Auto Mode: Improving Blocking Classifiers Against Malign Coding Agentscited by 0