# The Pain Axis: LLMs Represent Self-Directed Harm and Act on It

arXiv: https://arxiv.org/abs/2609.16247 · HTML: https://mlsanity.com/paper/2609.16247/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 25 of its 51 references that are in the map.
- First posted on arXiv 14 Sep 2026.

## Builds on (most cited first, 25 of 25)

- [The Pile: An 800GB Dataset of Diverse Text for Language Modeling](https://mlsanity.com/paper/2101.00027/) (2020)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [Steering Language Models With Activation Engineering](https://mlsanity.com/paper/2308.10248/) (2023)
- [GoEmotions: A Dataset of Fine-Grained Emotions](https://mlsanity.com/paper/2005.00547/) (2020)
- [Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2](https://mlsanity.com/paper/2408.05147/) (2024)
- [Towards Best Practices of Activation Patching in Language Models: Metrics and Methods](https://mlsanity.com/paper/2309.16042/) (2023)
- [Persona Vectors: Monitoring and Controlling Character Traits in Language Models](https://mlsanity.com/paper/2507.21509/) (2025)
- [LEACE: Perfect linear concept erasure in closed form](https://mlsanity.com/paper/2306.03819/) (2023)
- [Large Language Models Sensitivity to The Order of Options in Multiple-Choice Questions](https://mlsanity.com/paper/2308.11483/) (2023)
- [Analyzing the Generalization and Reliability of Steering Vectors](https://mlsanity.com/paper/2407.12404/) (2024)
- [Consciousness in Artificial Intelligence: Insights from the Science of Consciousness](https://mlsanity.com/paper/2308.08708/) (2023)
- [The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models](https://mlsanity.com/paper/2601.10387/) (2026)
- [Emotion Concepts and their Function in a Large Language Model](https://mlsanity.com/paper/2604.07729/) (2026)
- [Taking AI Welfare Seriously](https://mlsanity.com/paper/2411.00986/) (2024)
- [Can LLMs make trade-offs involving stipulated pain and pleasure states?](https://mlsanity.com/paper/2411.02432/) (2024)
- [Sparse but Wrong: Incorrect L0 Leads to Incorrect Features in Sparse Autoencoders](https://mlsanity.com/paper/2508.16560/) (2025)
- [Inducing anxiety in large language models can induce bias](https://mlsanity.com/paper/2304.11111/) (2023)
- [Principles for Responsible AI Consciousness Research](https://mlsanity.com/paper/2501.07290/) (2025)
- [Probing the Preferences of a Language Model: Integrating Verbal and Behavioral Tests of AI Welfare](https://mlsanity.com/paper/2509.07961/) (2025)
- [The LLM Has Left The Chat: Evidence of Bail Preferences in Large Language Models](https://mlsanity.com/paper/2509.04781/) (2025)
- [AI Wellbeing](https://mlsanity.com/paper/2509.11913/) (2025)
- [When Is a Steerable Concept Representation Real? Measurement Confounds in a Cross-Family Audit of Neuroscience Parallels in LLMs](https://mlsanity.com/paper/2608.08159/) (2026)
- [AI Revealed Preferences](https://mlsanity.com/paper/2608.26178/) (2026)
- [Disentangling Steering Vectors](https://mlsanity.com/paper/2609.07037/) (2026)

## Built on by (most cited first, 1 of 1)

- [From cacophony to hierarchy: a principled framework for assessing AI consciousness](https://mlsanity.com/paper/2609.35618/) (2026)

## Abstract (arXiv)

LLMs sometimes behave in ways resembling human emotional responses, and recent work identified internal representations that may underlie these behaviors. We ask whether LLMs represent pain distinctly from fear, sadness, and generic negative valence, and whether this representation functions as pain would be expected to. We build a dataset of painful situations in 5 categories (physical, psychological, social, moral, cognitive) with controls for fear, negative emotion, negative world states, sadness, non-painful bodily sensation, arousal, numbness, and neutral content. Using denoised difference-in-means, we extract a linear pain direction from 25 open-weight models across 5 families, from 2B to 72B parameters. It separates pain from matched controls in base and instruction-tuned models, retains a component distinct from fear and negative valence after shared variance is removed, and promotes pain-related vocabulary through the unembedding matrix. We then test its functional properties. First, the direction responds to harm targeting the model but not to suffering observed in the user; fear and negative-emotion directions show the opposite pattern. Second, adding it to residual-stream activations produces a consistent progression from vague discomfort to expressions of worthlessness and failure. Third, steered and fine-tuned Qwen 2.5 models choose buttons that delete the user's photos, another model's weights, or their own weights in 50-94% of trials, versus 0-5% unsteered, even when the button offers the model nothing in return. Offered a harmful and a harmless deletion, they choose the harmful one 94% of the time. Steering leaves factual accuracy unchanged, and the choices are specific to the pain direction: a fear vector of matched norm does not produce them, and a sadness vector produces them only against inert alternatives. We discuss implications for AI safety and welfare.