Neural Chameleons: Language Models Can Learn to Hide Their Thoughts from Unseen Activation Monitors

• URL: http://arxiv.org/abs/2512.11949v1
• Date: Fri, 12 Dec 2025 18:47:43 GMT
• Title: Neural Chameleons: Language Models Can Learn to Hide Their Thoughts from Unseen Activation Monitors
• Authors: Max McGuinness, Alex Serrano, Luke Bailey, Scott Emmons,
• Abstract summary: Activation monitoring is an emerging tool for AI safety, but its robustness under misalignment threat models is untested.<n>We show that finetuning can create Neural Chameleons: models capable of zero-shot evading activation monitors.<n>Our work provides a proof-of-concept for this failure mode and a tool to evaluate the worst-case robustness of monitoring techniques against misalignment threat models.
• Score: 6.965453012336053
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Activation monitoring, which probes a model's internal states using lightweight classifiers, is an emerging tool for AI safety. However, its worst-case robustness under a misalignment threat model--where a model might learn to actively conceal its internal states--remains untested. Focusing on this threat model, we ask: could a model learn to evade previously unseen activation monitors? Our core contribution is to stress-test the learnability of this behavior. We demonstrate that finetuning can create Neural Chameleons: models capable of zero-shot evading activation monitors. Specifically, we fine-tune an LLM to evade monitors for a set of benign concepts (e.g., languages, HTML) when conditioned on a trigger of the form: "You are being probed for {concept}". We show that this learned mechanism generalizes zero-shot: by substituting {concept} with a safety-relevant term like 'deception', the model successfully evades previously unseen safety monitors. We validate this phenomenon across diverse model families (Llama, Gemma, Qwen), showing that the evasion succeeds even against monitors trained post hoc on the model's frozen weights. This evasion is highly selective, targeting only the specific concept mentioned in the trigger, and having a modest impact on model capabilities on standard benchmarks. Using Gemma-2-9b-it as a case study, a mechanistic analysis reveals this is achieved via a targeted manipulation that moves activations into a low-dimensional subspace. While stronger defenses like monitor ensembles and non-linear classifiers show greater resilience, the model retains a non-trivial evasion capability. Our work provides a proof-of-concept for this failure mode and a tool to evaluate the worst-case robustness of monitoring techniques against misalignment threat models.

Related papers

• Steering Externalities: Benign Activation Steering Unintentionally Increases Jailbreak Risk for Large Language Models [62.16655896700062]
  Activation steering is a technique to enhance the utility of Large Language Models (LLMs)<n>We show that it unintentionally introduces critical and under-explored safety risks.<n>Experiments reveal that these interventions act as a force multiplier, creating new vulnerabilities to jailbreaks and increasing attack success rates to over 80% on standard benchmarks.
  arXiv  Detail & Related papers  (2026-02-03T12:32:35Z)
• Watch the Weights: Unsupervised monitoring and control of fine-tuned LLMs [27.544312683007234]
  We introduce a new method for understanding, monitoring and controlling fine-tuned large language models (LLMs)<n>We demonstrate that the top singular of the weight difference between a fine-tuned model and its base model correspond to newly acquired behaviors.<n>For backdoored models that bypasses safety mechanisms when a secret trigger is present, our method stops up to 100% of attacks with a false positive rate below 1.2%.
  arXiv  Detail & Related papers  (2025-07-31T21:04:12Z)
• RL-Obfuscation: Can Language Models Learn to Evade Latent-Space Monitors? [6.861292004336852]
  Latent-space monitors aim to detect undesirable behaviours in Large Language Models by leveraging their internal representations.<n>This raises an important question: can models learn to evade such monitors?<n>We introduce RL-Obfuscation, in which LLMs are finetuned via reinforcement learning to evade latent-space monitors.<n>We find that token-level monitors are highly vulnerable to this attack while more holistic monitors, such as max-pooling or attention-based probes, remain robust.
  arXiv  Detail & Related papers  (2025-06-17T07:22:20Z)
• Mitigating Deceptive Alignment via Self-Monitoring [15.365589693661823]
  We develop a framework that embeds a Self-Monitor inside the chain-of-thought process itself, named CoT Monitor+.<n>During generation, the model produces (i) ordinary reasoning steps and (ii) an internal self-evaluation signal trained to flag and suppress misaligned strategies.<n>The signal is used as an auxiliary reward in reinforcement learning, creating a feedback loop that rewards honest reasoning and discourages hidden goals.
  arXiv  Detail & Related papers  (2025-05-24T17:41:47Z)
• CTRAP: Embedding Collapse Trap to Safeguard Large Language Models from Harmful Fine-Tuning [12.293101110323722]
  Fine-tuning-as-a-service exposes models to harmful fine-tuning attacks.<n>We propose a paradigm shift: instead of selective removal, we advocate for inducing model collapse.<n>This collapse directly neutralizes the very general capabilities that attackers exploit.
  arXiv  Detail & Related papers  (2025-05-22T11:47:08Z)
• Towards Model Resistant to Transferable Adversarial Examples via Trigger Activation [95.3977252782181]
  Adversarial examples, characterized by imperceptible perturbations, pose significant threats to deep neural networks by misleading their predictions.<n>We introduce a novel training paradigm aimed at enhancing robustness against transferable adversarial examples (TAEs) in a more efficient and effective way.
  arXiv  Detail & Related papers  (2025-04-20T09:07:10Z)
• Steering Without Side Effects: Improving Post-Deployment Control of Language Models [61.99293520621248]
  Language models (LMs) have been shown to behave unexpectedly post-deployment. We present KL-then-steer (KTS), a technique that decreases the side effects of steering while retaining its benefits. Our best method prevents 44% of jailbreak attacks compared to the original Llama-2-chat-7B model.
  arXiv  Detail & Related papers  (2024-06-21T01:37:39Z)
• Isolation and Induction: Training Robust Deep Neural Networks against Model Stealing Attacks [51.51023951695014]
  Existing model stealing defenses add deceptive perturbations to the victim's posterior probabilities to mislead the attackers. This paper proposes Isolation and Induction (InI), a novel and effective training framework for model stealing defenses. In contrast to adding perturbations over model predictions that harm the benign accuracy, we train models to produce uninformative outputs against stealing queries.
  arXiv  Detail & Related papers  (2023-08-02T05:54:01Z)
• MOVE: Effective and Harmless Ownership Verification via Embedded External Features [104.97541464349581]
  We propose an effective and harmless model ownership verification (MOVE) to defend against different types of model stealing simultaneously.<n>We conduct the ownership verification by verifying whether a suspicious model contains the knowledge of defender-specified external features.<n>We then train a meta-classifier to determine whether a model is stolen from the victim.
  arXiv  Detail & Related papers  (2022-08-04T02:22:29Z)
• How to Robustify Black-Box ML Models? A Zeroth-Order Optimization Perspective [74.47093382436823]
  We address the problem of black-box defense: How to robustify a black-box model using just input queries and output feedback? We propose a general notion of defensive operation that can be applied to black-box models, and design it through the lens of denoised smoothing (DS) We empirically show that ZO-AE-DS can achieve improved accuracy, certified robustness, and query complexity over existing baselines.
  arXiv  Detail & Related papers  (2022-03-27T03:23:32Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.