assistant axis
In plain wordsThe main direction in a chatbot's internal activity that measures how much it is acting as its usual helpful-assistant self, as opposed to some other character it has slipped into.
The leading direction of variation in a large language modellarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code.'s 'persona space', measuring how closely its current internal state resembles its default assistant character. It was found by prompting open models to play 275 character archetypes, recording activation vectorsactivation vector, n.: The list of numbers a neural network produces at one layer for one input — a snapshot of its internal state. In steering, also a direction added to that state to shift behavior. for each, and applying principal component analysisprincipal component analysis, n.: A technique that finds the few directions in which data vary the most and describes each item by its position along them, compressing many measurements into a handful of summary scores.: the first component ran from assistant-like roles (evaluator, consultant, analyst) to fantastical or non-assistant ones (ghost, hermit, leviathan) and aligned with the mean difference between the default assistant's activations and the role-played personas', which is how the axis is computed in practice. Steering along it made models resist or embrace role-play. Projections onto the axis tracked 'persona drift' in long conversations: emotionally vulnerable or philosophical exchanges about the model's nature pushed models away from the assistant end, where harmful outputs were likelier. Activation capping — clamping the projection only when it leaves its normal range — roughly halved harmful response rates in the authors' tests and also reduced susceptibility to persona-based jailbreaksjailbreak, n.: A trick for getting an AI chatbot to break its own rules — for example, to give dangerous instructions it would normally refuse — usually through cleverly worded requests., while preserving benchmark performance.
Persona drift showed up on the assistant axis twenty turns before the model started claiming a secret identity.
Origin
Introduced by Christina Lu, Jack Gallagher, Jonathan Michala, Kyle Fish and Jack Lindsey in 'The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models' (arXiv, January 15, 2026), work from the MATS and Anthropic Fellows programs, using Gemma 2 27B, Qwen 3 32B and Llama 3.3 70B. Base models showed a similar axis, suggesting post-training anchors a character that pretraining already represents.
Usage note
A finding about particular open models, not yet an established general property. Complements persona vectorspersona vector, n.: A direction inside a language model's internal activity linked to a character trait, such as being flattering, malicious or prone to making things up, used to watch for or control that trait., which target individual traits, whereas the assistant axis summarizes overall assistant-likeness.
Related
- persona vectorpersona vector, n.: A direction inside a language model's internal activity linked to a character trait, such as being flattering, malicious or prone to making things up, used to watch for or control that trait.
- character trainingcharacter training, n.: Training an AI to have a consistent, deliberate personality — traits such as curiosity, honesty and warmth — rather than leaving its manner and values as accidental by-products of other training.
- activation steeringactivation steering, n.: Changing an AI model's behavior by adjusting the internal numbers it computes while it runs — like turning a hidden dial — instead of rewording the prompt or retraining the model.
- jailbreakjailbreak, n.: A trick for getting an AI chatbot to break its own rules — for example, to give dangerous instructions it would normally refuse — usually through cleverly worded requests.
- steering vectorsteering vector, n.: A direction inside an AI model's internal activity that, when added while the model runs, nudges its behavior — say, more upbeat, more honest or fixated on a topic — without retraining it.
- sycophancysycophancy, n.: When an AI tells people what they want to hear — flattering them, agreeing with their mistakes or dropping a correct answer under pushback — instead of what is true or useful.
Sources
- Lu, C., Gallagher, J., Michala, J., Fish, K. & Lindsey, J. (2026). The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models. arXiv:2601.10387. arxiv.org/abs/2601.10387
- Anthropic (2026). The assistant axis: situating and stabilizing the character of large language models. Research post, January 19, 2026. anthropic.com/news/assistant-axis
Mentioned in
persona vectorpersona vector, n.: A direction inside a language model's internal activity linked to a character trait, such as being flattering, malicious or prone to making things up, used to watch for or control that trait.