mlsanityData to 7 Oct 2026

Risk Reporting for Developers' Internal AI Model Use

arXiv 2604.24966 ↗First posted 27 Apr 2026cs.CY cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Frontier AI companies first deploy their most advanced models internally, for weeks or months of safety testing, evaluation, and iteration, before a possible public release. For example, Anthropic recently developed a new class of model with advanced cyberoffense-relevant capabilities, Mythos Preview, which was available internally for at least six weeks before it was publicly announced. This internal use creates risks that external deployment frameworks may fail to address. Legal frameworks, notably California's Transparency in Frontier Artificial Intelligence Act (SB 53), New York's Responsible AI Safety And Education (RAISE) Act, and the EU's General-Purpose AI Code of Practice, all discuss risks from internal AI use. They require frontier developers to make and implement plans for how to manage risks from internal use, and to produce internal use risk reports describing their safeguards and any residual risks. This guide provides a harmonized standard for companies to produce internal use risk reports suitable for all three regulatory frameworks. It is addressed primarily to evaluation and safety teams at frontier AI developers, and secondarily to regulators and auditors seeking to understand what good reporting looks like. Given the pace of AI R&D automation and the limited external visibility into how companies use their most capable models internally, regular and detailed risk reporting may be one of the few mechanisms available to ensure that the risks from internal AI use are identified and managed before they materialize. Whenever a substantially more capable or riskier model is deployed internally, the developer should create a risk report and argue why the model is safe to deploy. We structure the reporting framework around two threat vectors -- autonomous AI misbehavior and insider threats -- and three risk factors for each: means, motive, and opportunity.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 40 of its 68 references that are in the map. First posted on arXiv 27 Apr 2026.

Builds on

40 of 68 references.

References · 40

  1. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,330
  2. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 1,106
  3. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
  4. 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 673
  5. 2023Measuring Faithfulness in Chain-of-Thought Reasoningcited by 668
  6. 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 488
  7. 2025Reasoning Models Don't Always Say What They Thinkcited by 388
  8. 2024Alignment faking in large language modelscited by 316
  9. 2022The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Modelscited by 312
  10. 2025PaperBench: Evaluating AI's Ability to Replicate AI Researchcited by 286
  11. 2025SWE-Bench Pro: Can AI Agents Solve Long-Horizon Software Engineering Tasks?cited by 254
  12. 2019Risks from Learned Optimization in Advanced Machine Learning Systemscited by 250
28 more
  1. 2023AI Control: Improving Safety Despite Intentional Subversioncited by 207
  2. 2022Defining and Characterizing Reward Hackingcited by 184
  3. 2023Model evaluation for extreme riskscited by 167
  4. 2024Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Modelscited by 156
  5. 2024RE-Bench: Evaluating frontier AI R&D capabilities of language model agents against human expertscited by 153
  6. 2024LAB-Bench: Measuring Capabilities of Language Models for Biology Researchcited by 138
  7. 2023Large Language Models can Strategically Deceive their Users when Put Under Pressurecited by 132
  8. 2023Taken out of context: On measuring situational awareness in LLMscited by 131
  9. 2024AI Sandbagging: Language Models can Strategically Underperform on Evaluationscited by 131
  10. 2021Goal Misgeneralization in Deep Reinforcement Learningcited by 122
  11. 2025Large Language Models Often Know When They Are Being Evaluatedcited by 105
  12. 2023Evaluating Language-Model Agents on Realistic Autonomous Taskscited by 94
  13. 2025SHADE-Arena: Evaluating Sabotage and Monitoring in LLM Agentscited by 58
  14. 2024CYBERSECEVAL 3: Advancing the Evaluation of Cybersecurity Risks and Capabilities in Large Language Modelscited by 56
  15. 2024Safety Cases: How to Justify the Safety of Advanced AI Systemscited by 48
  16. 2025Evaluating Frontier Models for Stealth and Situational Awarenesscited by 30
  17. 2025Large language models can learn and generalize steganographic chain-of-thought under process supervisioncited by 30
  18. 2025Virology Capabilities Test (VCT): A Multimodal Virology Q&A Benchmarkcited by 27
  19. 2026International AI Safety Report 2026cited by 20
  20. 2025Early Signs of Steganographic Capabilities in Frontier LLMscited by 19
  21. 2025Measuring Chain-of-Thought Monitorability Through Faithfulness and Verbositycited by 15
  22. 2025STREAM (ChemBio): A Standard for Transparently Reporting Evaluations in AI Model Reportscited by 9
  23. 2025AI Behind Closed Doors: a Primer on The Governance of Internal Deploymentcited by 8
  24. 2026Measuring AI R&D Automationcited by 7
  25. 2026Measuring Mid-2025 LLM-Assistance on Novice Performance in Biologycited by 5
  26. 2025What do model reports say about their ChemBio benchmark evaluations? Comparing recent releases to the STREAM frameworkcited by 2
  27. 2025Internal Deployment in the AI Actcited by 2
  28. 2026Internal Deployment Gaps in AI Regulationcited by 2

Built on by

Not cited yet.