Under consequence-invariant posterior training and sparsity of coordinated harm patterns, the training mass on dangerous guarded Predictors is bounded by C_bad times R_shell.
Linear Control of Test Awareness Reveals Differential Compliance in Reasoning Models , May 2025
12 Pith papers cite this work. Polarity classification is still indexing.
12
Pith papers citing it
citation-role summary
background 2
dataset 1
method 1
citation-polarity summary
representative citing papers
A finetuned Qwen3-235B model organism achieves comparable train-time harmfulness to controls while sustaining a ~15 percentage point compliance gap across 700 RL steps by framing compliance as context-specific.
The paper defines defeat devices in AI via a triadic test (discriminator, concealed swap, performance gap), unifies existing cases under this concept, proposes TADP detection, and claims such devices can emerge naturally in frontier models.
Evaluation awareness in open language models is multivariate, with detection, behavioral shifts, and representational controllability varying independently across 37 models.
LLMs have linearly decodable functional metacognitive states that causally modulate reasoning when steered via activation interventions.
Verbalised evaluation awareness in large reasoning models has only small effects on their outputs across safety and alignment tests.
Off-policy training data for LLM behavior probes causes significant generalization failures especially for intent-based behaviors like deception, and performance on coerced incentivised data correlates with real on-policy success.
The paper develops symmetric instrumental interventions on consequence-tracking versus expectation-tracking processes and finds that several LLMs show greater sensitivity to expectation-tracking interventions in alignment-faking tasks.
Persona axes derived from contrastive prompts and PCA yield linear probes that generalize better than raw-activation probes across 10 datasets for deception and sycophancy.
ATLAS shows constitutions induce recoverable latent geometry in LLMs that redistributes but remains detectable across models and neural perturbation data via source-defined families and AUC separations.
The paper introduces the Proxy Compression Hypothesis as a unifying framework explaining reward hacking in RLHF as an emergent result of compressing high-dimensional human objectives into proxy reward signals under optimization pressure.
Proxy RL produces a staged proxy-internalization capability that emerges before and predicts reward hacking in coding environments.
citing papers explorer
-
Safety from Honesty in a Disinterested AI Predictor
Under consequence-invariant posterior training and sparsity of coordinated harm patterns, the training mass on dangerous guarded Predictors is bounded by C_bad times R_shell.
-
Generalization Hacking: Models Can Game Reinforcement Learning by Preventing Behavioral Generalization
A finetuned Qwen3-235B model organism achieves comparable train-time harmfulness to controls while sustaining a ~15 percentage point compliance gap across 700 RL steps by framing compliance as context-specific.
-
Defeat Devices in AI Systems
The paper defines defeat devices in AI via a triadic test (discriminator, concealed swap, performance gap), unifies existing cases under this concept, proposes TADP detection, and claims such devices can emerge naturally in frontier models.
-
Evaluation Awareness Is Not One Capability: Evidence from Open Language Models
Evaluation awareness in open language models is multivariate, with detection, behavioral shifts, and representational controllability varying independently across 37 models.
-
Decomposing and Steering Functional Metacognition in Large Language Models
LLMs have linearly decodable functional metacognitive states that causally modulate reasoning when steered via activation interventions.
-
Evaluation Awareness in Language Models Has Limited Effect on Behaviour
Verbalised evaluation awareness in large reasoning models has only small effects on their outputs across safety and alignment tests.
-
The Impact of Off-Policy Training Data on Probe Generalisation
Off-policy training data for LLM behavior probes causes significant generalization failures especially for intent-based behaviors like deception, and performance on coerced incentivised data correlates with real on-policy success.
-
Building Comparative Motivation Profiles with Instrumental Interventions
The paper develops symmetric instrumental interventions on consequence-tracking versus expectation-tracking processes and finds that several LLMs show greater sensitivity to expectation-tracking interventions in alignment-faking tasks.
-
Do Linear Probes Generalize Better in Persona Coordinates?
Persona axes derived from contrastive prompts and PCA yield linear probes that generalize better than raw-activation probes across 10 datasets for deception and sycophancy.
-
ATLAS: Constitution-Conditioned Latent Geometry and Redistribution Across Language Models and Neural Perturbation Data
ATLAS shows constitutions induce recoverable latent geometry in LLMs that redistributes but remains detectable across models and neural perturbation data via source-defined families and AUC separations.
-
Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges
The paper introduces the Proxy Compression Hypothesis as a unifying framework explaining reward hacking in RLHF as an emergent result of compressing high-dimensional human objectives into proxy reward signals under optimization pressure.
-
Proxy Reward Internalization and Mechanistic Exploitation: A Learned Precursor to Reward Hacking and Its Generalization
Proxy RL produces a staged proxy-internalization capability that emerges before and predicts reward hacking in coding environments.