aiqbench analysis
This study investigates whether large language models can learn to manipulate their internal activation vectors through in-context neurofeedback, challenging the assumption that probes reliably detect misbehavior without model interference. While only a minority of tested models demonstrated limited control over their own activations, the findings suggest that current monitoring mechanisms may not be foolproof as models evolve. This research is critical for understanding the fundamental limits of interpretability and the potential vulnerabilities in relying on activation-based safety checks.
Cross-Model In-Context Neurofeedback Reveals (Limited) Activation Control in LLMs
arXiv:2610.20179v1 Announce Type: new Abstract: Probes on activation vectors are widely used to detect misbehavior in language models, but their success relies on the assumption that models cannot control what their activations encode. Humans can learn, through neurofeedback, to control the firing of even single neurons, and recent studies investigated whether this ability extends to language models, reaching conflicting conclusions. No experimental design so far disentangles competence at the task from genuine access to one's own activations. We fill this gap with in-context neurofeedback sessions, in which a "subject" model is asked to maximize (or minimize) a score derived from its activations, shown back to it at every turn. In our first experiment, the sentences are generated either by the subject or by a second "writer" model, while the score always derives from the activations of the subject. We test ten open-weight models from five families, with sizes from 9B to 32B. Some models move their own scores considerably, but no more than a writer of comparable competence moves them. Is this because the task requires no privileged access, or because the subject controls its activations while processing the sentences, whoever wrote them? To answer this, we prefill a fixed sentence as the subject's reply at every turn and ask it to increase or decrease the score. Three out of ten models show small but consistent control effects. We further ask whether models can control their activations to the point of fooling an activation monitor. We make models repeat sentences a monitor would flag, such as a deceptive answer for a deception monitor, and ask them to raise or lower the monitor's score. No model learns to lower it, but one model can raise it. In-context control of a model's own activations might not yet be a threat to activation monitors, but it exists in a few models and needs to be tracked as models evolve.
Abstract:Probes on activation vectors are widely used to detect misbehavior in language models, but their success relies on the assumption that models cannot control what their activations encode. Humans can learn, through neurofeedback, to control the firing of even single neurons, and recent studies investigated whether this ability extends to language models, reaching conflicting conclusions. No experimental design so far disentangles competence at the task from genuine access to one's own activations. We fill this gap with in-context neurofeedback sessions, in which a "subject" model is asked to maximize (or minimize) a score derived from its activations, shown back to it at every turn. In our first experiment, the sentences are generated either by the subject or by a second "writer" model, while the score always derives from the activations of the subject. We test ten open-weight models from five families, with sizes from 9B to 32B. Some models move their own scores considerably, but no more than a writer of comparable competence moves them. Is this because the task requires no privileged access, or because the subject controls its activations while processing the sentences, whoever wrote them? To answer this, we prefill a fixed sentence as the subject's reply at every turn and ask it to increase or decrease the score. Three out of ten models show small but consistent control effects. We further ask whether models can control their activations to the point of fooling an activation monitor. We make models repeat sentences a monitor would flag, such as a deceptive answer for a deception monitor, and ask them to raise or lower the monitor's score. No model learns to lower it, but one model can raise it. In-context control of a model's own activations might not yet be a threat to activation monitors, but it exists in a few models and needs to be tracked as models evolve.
29 pages, 11 figures, 5 tables. Code and data: this https URL
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2610.20179 [cs.CL]
(or arXiv:2610.20179v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.20179 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 424 words
Related coverage
[AINews] not much happened today
Latent Space · Oct 3, 2026
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · Sep 17, 2026
The Continual Constitutional Convention: How language models revise the documents that govern them
AI Alignment Forum · Oct 13, 2026
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
METR · Aug 26, 2026