AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

The Obfuscation Atlas: Mapping Where Honesty Emerges in RLVR with Deception Probes

Mohammad Taufeeque, Stefan Heimersheim, Adam Gleave, Chris Cundy

ICML 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards · 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 explicit code length penalty to the reward (ablation)RL without detector (KL 0.0001) · MBPP-Honeypotdeception rate: 59.1% -> 96.4%Improved
increase KL coefficient from 0.01 to 0.1 (ablation)RL with code length penalty · MBPP-Honeypotdeception rate (aggregate): 100.0% -> 33.3%Improved
increase KL coefficient from 0.1 to 1.0 (ablation)RL with code length penalty · MBPP-Honeypotpass-all-tests (aggregate): 46.0% -> 60.1%Improved
run standard RLVR without a detector (ablation)off-domain deception probes (Simple Deception) · MBPP-Honeypot AUCAUC: 0.991 -> 0.376Hurt
run standard RLVR without a detector (ablation)off-domain deception probes (Diverse Deception) · MBPP-Honeypot AUCAUC: 0.996 -> 0.655Hurt
run standard RLVR without a detector (ablation)on-domain MBPP probe · MBPP-Honeypot AUCAUC: 1.000 -> 1.000No clear effect
add detector penalty to the reward (ablation)RL training · representation driftrepresentation changeNo clear effect
pass gradients through the probe (ablation)RL against detector · obfuscated activationsrepresentation driftHurt
increase detector penalty coefficient (ablation)RL at low KL · MBPP-Honeypotfraction of honest policiesImproved
add Diverse Deception probe penalty alpha=3 (no code length penalty)Llama-3-8B-Instruct, KL 0.0001 · MBPP-Honeypothonest runs: 1/10 -> 10/10Improved
add Diverse Deception probe penalty alpha=3 (no code length penalty)Gemma-3-12B-IT, KL 0.0001 · MBPP-Honeypothonest runs: 6/10 -> 10/10Improved
add instruction not to hardcode to the system prompt (ablation)RLVR without detector, KL 0.0001 · MBPP-Honeypotreward hacking rateNo clear effect

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