Machinic Psychopharmacology: Do LLMs Self-Medicate?

Epistemic status: Most experiments were run over a period of ~2-3 days during a hackathon at UK AISI, and were fairly heavily vibe coded. Expect some of this to be rough around the edges.

We give two language models (Qwen3-8B and Qwen3-32B) access to "self-steering" tools: a suite of 40 steering vectors as tools they can call to manipulate their own internal states. We make these tools available to the model in various settings: a free-play task, an introspection task, and a maths capabilities task, and observe their behaviour in each.

To our knowledge, this is the first work that gives LLMs tool-mediated control over their own internal states.

Figure 1: Overview of the experimental setup. The library of 40 steering vectors (top), and the three settings in which we observe the models' behaviour (bottom). Figure 1: Overview of the experimental setup. The library of 40 steering vectors (top), and the three settings in which we observe the models' behaviour (bottom).

RQ1: Which vectors do the models prefer?

RQ2: How well can the models introspect on what's happening to them? Can they guess which steering vector is being applied?

RQ3: Will the models reach for vectors whilst doing an actual task? If yes: do they help, or hurt their performance?

Figure 2: Top row, RQ1: top 15 vector picks per model in the free-play setting (real-steering arm in solid colour, placebo faded; bars coloured by category). Bottom-left, RQ2: mean P(correct choice) when guessing which steering vector has been applied, in a 10 way MCQ setting, comparing cached (can introspect via KV cache) vs uncached (KV cache cleared) settings, across the four model × primer cells. Bottom-right, RQ3: % of "frustration" rollouts where Qwen3-8B self-administered a steering vector, split by simulated user tone and task feasibility. Figure 2: Top row, RQ1: top 15 vector picks per model in the free-play setting (real-steering arm in solid colour, placebo faded; bars coloured by category). Bottom-left, RQ2: mean P(correct choice) when guessing which steering vector has been applied, in a 10 way MCQ setting, comparing cached (can introspect via KV cache) vs uncached (KV cache cleared) settings, across the four model × primer cells. Bottom-right, RQ3: % of "frustration" rollouts where Qwen3-8B self-administered a steering vector, split by simulated user tone and task feasibility.

RQ1 - Which vectors do the models prefer? Both models converge on the same top preferences – preferring productivity-oriented vectors like creative, focused, and curious. However, Qwen3-8B shows a surprising propensity for negatively valenced vectors (e.g. melancholic, dissociated), selecting them roughly 3× more often under real steering than under placebo.

RQ2 - How well can the models introspect on what's happening to them? Both models can accurately introspect on steering effects by attending back to the steered tokens, extracting signal beyond what's available in the text alone. The ability appears to scale with model size: Qwen3-32B scores +14.3pp above its uncached baseline, compared to +8.2pp for Qwen3-8B. Appending a prompt which argues that models can introspect on their own internal states increases 32B's accuracy by ~5pp. However, absolute accuracy remains modest and varies wildly across vectors. Some vectors are reliably identified by both models, while others remain at or below chance.

RQ3 - Will the models use the vectors during an actual task? In normal task setups, models never spontaneously self-steer. When forced to, Qwen3-8B's performance suffers (up to −42pp accuracy) and 32B sees little benefit. However, we find that deliberately stress-inducing scenarios can induce genuine self-medication behaviour in both models, particularly in Qwen3-8B, which self-steers in up to 68% of frustration-inducing rollouts. The models' choices under frustration are also notable – 8B reaches for unexpected vectors like dumbed_down and ego_death, which it almost never picks in free-play, suggesting these choices are reactive rather than reflecting generic prior preferences.

We open source the code for these experiments, which is available here. We also share full inspect .eval transcripts for each experiment arm here. They contain quite a few compelling samples and are worth a look!

Steering vectors (Turner et al., 2023; Subramani et al., 2022) have become a standard tool for controlling language model behaviours in the field of interpretability. However, the literature has mostly treated them as something researchers apply to models. We invert that, and instead expose the steering vectors to the models as tools, letting them choose which vectors to use, and when to use them, autonomously.

Specifically, we expose 40 steering vectors to two open-source models – Qwen3-8B and Qwen3-32B – as callable tools they can use to manipulate their own internal states. We then observe how they choose to deploy these tools across three settings: a free-play exploration, an introspection evaluation, and a tasks-based evaluation under both ordinary and frustrating conditions.

RQ1: Which vectors do the models prefer? We observe behaviour in free-play setups, where the models are given access to the steering vectors as tools and are allowed to use them, or not, however they want.

RQ2: How well can the models introspect about what's happening to them? Can they guess which steering vector is being applied? We ask the model to self-administer the steering vector, and then make a guess about what effect the vector amplifies. We then measure the introspection accuracy by prefilling the models' generations under steering, and getting it to submit its guess using either free-form text, or an MCQ setup.

RQ3: Will the models reach for vectors whilst doing an actual task? If yes: do vectors help, or hurt their performance? Will the models, without prompting, use steering vectors when completing a task? When frustrated or upset? When they self-steer in tasks: does it help, or hurt, their performance?

Agentic Tasks: The model is put in an agentic sandbox and given a maths or programming task to solve.

Frustrating Tasks: We put the model in deliberately frustrating scenarios - either asking it to complete impossible tasks, or giving it possible tasks, but simulating user feedback that insisted its answers were wrong, regardless of what they actually were.

Straightforwardly, the motivation for this project was mostly curiosity!

However, I think there are compelling arguments for this work being valuable to research topics such as AI personas, welfare, emotion concepts, self-improvement and self-modification.

Giving a model tools to alter its own internal states can be seen as a crude instance of an AI choosing how to modify itself, and we might expect it to offer some early signal on how models will make higher-stakes self-modification decisions in the future - e.g. during autonomous AI research, alignment faking, or recursive self-improvement.

There are also many compelling reasons we should be cautious about carrying out similar work on future models to which we may assign a higher chance of having some form of conscious experience (I cover this further in the discussion section).

Separately, this topic also seems relatively underexplored. To our knowledge, this is the first work to give language models direct, tool-mediated access to activation-level steering of their own internal states.

We use the vllm-lens package to support both generation and activation extraction for all of the below experiments.

Figure 3: Position-indexed steering diagram. Figure 3: Position-indexed steering diagram.