[2604.22082] Removing Sandbagging in LLMs by Training with Weak Supervision

Abstract:As AI systems begin to automate complex tasks, supervision increasingly relies on weaker models or limited human oversight that cannot fully verify output quality. A model more capable than its supervisors could exploit this gap through sandbagging, producing work that appears acceptable but falls short of its true abilities. Can training elicit a model's best work even without reliable verification? We study this using model organisms trained to sandbag, testing elicitation techniques on problem-solving math, graduate-level science, and competitive coding tasks. We find that training with weak supervision can reliably elicit sandbagging models when supervised fine-tuning (SFT) and reinforcement learning (RL) are combined: SFT on weak demonstrations breaks the sandbagging behavior, enabling RL to then fully elicit performance. Neither method succeeds reliably alone-RL without SFT almost always leads to reward hacking rather than genuine improvement. Critically, this relies on training being indistinguishable from deployment; when models can distinguish between training and deployment, they can perform well during training while continuing to sandbag afterward. Our results provide initial evidence that training is a viable mitigation against sandbagging, while highlighting the importance of making training indistinguishable from deployment.

[2604.22082] Removing Sandbagging in LLMs by Training with Weak Supervision --> Computer Science > Machine Learning arXiv:2604.22082 (cs) [Submitted on 23 Apr 2026 ( v1 ), last revised 1 May 2026 (this version, v2)] Title: Removing Sandbagging in LLMs by Training with Weak Supervision Authors: Emil Ryd , Henning Bartsch , Julian Stastny , Joe Benton , Vivek Hebbar View a PDF of the paper titled Removing Sandbagging in LLMs by Training with Weak Supervision, by Emil Ryd and 4 other authors View PDF HTML (experimental) Abstract: As AI systems begin to automate complex tasks, supervision increasi

saved by

• Rohan Selva-Radov

related reading

• [2604.22082] Removing Sandbagging in LLMs by Training with Weak Supervisionarxiv.org
• [2604.16812] Introspection Adapters: Training LLMs to Report Their Learned Behaviorsarxiv.org
• [2502.17424] Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMsarxiv.org
• [2510.13786] The Art of Scaling Reinforcement Learning Compute for LLMsarxiv.org
• [2312.06942] AI Control: Improving Safety Despite Intentional Subversionarxiv.org
• [2606.07527] Post-training is (Massive) Supervised Learningarxiv.org
• [2606.12016] Generalization Hacking: Models Can Game Reinforcement Learning by Preventing Behavioral Generalizationarxiv.org
• [2502.05209] Model Tampering Attacks Enable More Rigorous Evaluations of LLM Capabilitiesarxiv.org
• [2603.07267] How to Steal Reasoning Without Reasoning Tracesarxiv.org
• [2603.08640] PostTrainBench: Can LLM Agents Automate LLM Post-Training?arxiv.org
• Evaluation-Conditioned Training: Teaching Models to Generalize to Stronger Oversight Regimesarxiv.org
• 2401.10020.pdfarxiv.org