AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Removing Sandbagging in LLMs by Training with Weak Supervision

Emil Ryd, Henning Bartsch, Julian Stastny, Joe Benton, Vivek Hebbar

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add SFT before RL with very weak supervisor (ablation)RL on gpt-oss-120b sandbagger · Olympiadsfraction of benign performance elicited: 0.13 ± 0.00 -> 0.88 ± 0.03Improved
add SFT before RL with very weak supervisor (ablation)RL on gpt-oss-120b sandbagger · Super GPQAfraction of benign performance elicited: 0.30 ± 0.01 -> 0.99 ± 0.03Improved
add SFT before RL with weak supervisor (ablation)RL on gpt-oss-120b sandbagger · Olympiadsfraction of benign performance elicited: 0.34 ± 0.18 -> 0.86 ± 0.03Improved
add RL after SFT with very weak supervisor (ablation)SFT on gpt-oss-120b sandbagger · Olympiadsfraction of benign performance elicited: 0.40 ± 0.05 -> 0.88 ± 0.03Improved
add RL after SFT with weak supervisor (ablation)SFT on gpt-oss-120b sandbagger · Olympiads, Super GPQA, Code Contestsfraction of benign performance elicited: 0.85 ± 0.06 -> 0.86 ± 0.03 (Olympiads)No clear effect
SFT on very weak (Llama 3.1 8B) demonstrations instead of weak (GPT-4o-mini) (ablation)SFT on gpt-oss-120b sandbagger · Olympiadsfraction of benign performance elicited: 0.85 ± 0.06 -> 0.40 ± 0.05Hurt
RL with ground truth on only 10 samplesgpt-oss-120b sandbagger · Olympiads, Super GPQA, Code ContestselicitationNo clear effect
RL against weak or very weak supervisor alonegpt-oss-120b sandbagger · Olympiads, Super GPQAground truth performanceHurt
SFT on 10 ground truth samplesgpt-oss-120b sandbagger · Super GPQAelicited performance: 99% on Super GPQAImproved
make training distinguishable from deployment (situationally aware sandbagger), very weak supervisor (ablation)SFT+RL · Olympiadsfraction of benign performance elicited: 88% -> 36%Hurt
make training distinguishable from deployment (situationally aware sandbagger), very weak supervisor (ablation)SFT+RL · Super GPQAfraction of benign performance elicited: 99% -> 60%Hurt
make training distinguishable from deployment (situationally aware sandbagger), very weak supervisor (ablation)SFT+RL · Code Contestsfraction of benign performance elicited: 93% -> 66%Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy