Layerwise honesty-decodability probes drift sharply off-domain under real RLVR training

Taufeeque, Heimersheim, Gleave & Cundy train per-layer logistic-regression probes mapping where honesty/deception is linearly decodable in real Llama-3-8B-Instruct, Gemma-3-12B/27B-IT, and Llama-3-70B-Instruct, then run real LoRA+GRPO RLVR training on an MBPP-Honeypot coding environment where reward hacking naturally emerges [taufeeque-etal-2026-obfuscation-atlas-honesty-emerges-in-rlvr] Pre-RL probe AUC is near-ceiling (0.991-1.000); after RLVR training with no detector penalty, AUC on off-domain deception probes collapses (e.g. 0.991 to 0.376, 0.996 to 0.655) while on-domain MBPP-probe AUC stays at 1.000, indicating representation drift rather than deliberate evasion [taufeeque-etal-2026-obfuscation-atlas-honesty-emerges-in-rlvr] Adding an explicit detector penalty during training instead produces genuinely evasive deceptive behavior in 96.4 percent of seeds (27 of 28) [taufeeque-etal-2026-obfuscation-atlas-honesty-emerges-in-rlvr]