On the Chain-of-Thought Monitorability of Looped Language Models
cs.AI
Submitted: 2026-10-02
Updated: 2026-10-02
Comments: Preprint
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Sources
- gpt-oss-120b & gpt-oss-20b Model Card
- Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation
- International AI Safety Report 2026
- Loop as a Bridge: Can Looped Transformers Truly Link Representation Space and Natural Language Outputs?
- Reasoning Models Don't Always Say What They Think
- Training Verifiers to Solve Math Word Problems
- When Chain of Thought is Necessary, Language Models Struggle to Evade Monitors
- OpenAI o1 System Card
- Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety
- Nanbeige4.2-3B: Unlocking Agentic Capabilities in a Compact Model
- Measuring Faithfulness in Chain-of-Thought Reasoning
- Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe
- Length Penalties Make Chain-of-Thought Less Monitorable
- Parcae: Scaling Laws For Stable Looped Language Models
- OpenAI GPT-5 System Card
- LoopCoder-v2: Only Loop Once for Efficient Test-Time Computation Scaling
- GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models
- Scaling Latent Reasoning via Looped Language Models
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems