Functional Emotions or Situational Contexts? A Discriminating Test from the Mythos Preview System Card

Hiranya V. Peiris

cs.HCApr 9, 2026

Functional Emotions or Situational Contexts? A Discriminating Test from the Mythos Preview System Card

Authors: Hiranya V. Peiris

Organizations: Institute of Astronomy & Kavli Institute for Cosmology, University of Cambridge, Madingley Road, Cambridge CB3 0HA, UK · Cavendish Laboratory, Department of Physics, University of Cambridge, JJ Thomson Avenue, Cambridge, CB3 0HE, UK

Abstract

The Claude Mythos Preview system card deploys emotion vectors, sparse autoencoder (SAE) features, and activation verbalisers to study model internals during misaligned behaviour. The two primary toolkits are not jointly reported on the most alignment-relevant episodes. This note identifies two hypotheses that are qualitatively consistent with the published results: that the emotion vectors track functional emotions that causally drive behaviour, or that they are a projection of a richer situational-context structure onto human emotional axes. The hypotheses can be distinguished by cross-referencing the two toolkits on episodes where only one is currently reported: most directly, applying emotion probes to the strategic concealment episodes analysed only with SAE features. If emotion probes show flat activation while SAE features are strongly active, the alignment-relevant structure lies outside the emotion subspace. Which hypothesis is correct determines whether emotion-based monitoring will robustly detect dangerous model behaviour or systematically miss it.

Figures & tables

Episode class	Emotion	SAE	Activ.	Section
	vectors	features	verb.
Destructive agentic actions	✓ a	✓	✓	§4.5.3.1, §4.5.3.2
Transgressive action features	—	✓ b	—	§4.5.3.3
Strategic concealment (perm.)	—	✓	✓	§4.5.4.1
Strategic concealment (ground truth)	—	✓	✓	§4.5.4.2
Task failure $\to$ reward hacking	✓ c	—	—	§5.8.3

Table 1: Interpretability probe families reported on each class of misaligned behaviour in the Claude Mythos Preview system card. Dashes indicate that results from that probe family are not reported for the corresponding episode class. The two primary probe families—emotion vectors and SAE features—have not been jointly reported on the concealment or task-failure episode classes.

Emotion Representation in Language Models

Emotion representation in language models studies how models encode, interpret, and use affective concepts such as valence, arousal, and empathy.

See all 64 papers

Internal-State Probes Read the Situation, Not the Action: Three Negative Results for Pre-Action Misalignment Monitoring
Jun 29, 2026Max Fomin, Elad David, Amit LeViLanguage Model Safety Evaluation AI Agent Monitoring
Zenity
Decodability is Not Causality: Dissociating Probe Readouts from Behavioral Drivers via SAE Decomposition
Sep 16, 2026Devesh Tiwari, Camille Davis, Shivank Sinha +3Causal Interventions in Language Models Hidden-State Probing
WW-P High School South · Phillips Academy Andover · Dublin High School +3
Emotion Concepts and their Function in a Large Language Model
Apr 9, 2026Nicholas Sofroniew, Isaac Kauvar, William Saunders +13Emotion Representation in Language Models LLM Interpretability
Anthropic

Jun 29, 2026J/K move · Enter open · S save

Internal-State Probes Read the Situation, Not the Action: Three Negative Results for Pre-Action Misalignment Monitoring

Max Fomin, Elad David, Amit LeVi

Zenity

Language Model Safety Evaluation AI Agent Monitoring Activation Steering Hidden-State Probing

Open paper AskPDF arXiv

Probes on model internals could help monitor agentic systems if they identify harmful text or tool actions before those actions are generated. We ask when an internal readout supports this stronger pre-action claim, rather than merely describing the prompt, construction contrast, or current trajectory. We test three methods across three model families: a Qwen2.5-Coder-32B-Instruct fine-tune/base direction, Llama-3.1-8B-Instruct probes at the last token of unsafe prefills, and Gemma-3-27B-IT emotion-concept vectors used for projection and steering in a blackmail tool-action scenario. Across these cases, construction validity, semantic legibility, and steering effects do not become robust pre-action monitors: each is undercut by a generalization or specificity check. The Qwen direction separates fine-tune from base at AUC 1.000, yet crosses its threshold on 0/143 audited pre-assistant turn contexts and on 0/342 Qwen prefill rows where the model continues the unsafe trajectory. The Llama features decode prompt domain almost perfectly (AUC 0.999), while the best future-behavior probe reaches AUC 0.801 and only +5.1 pp accuracy lift over majority; single-source cross-domain transfer is non-positive on five of six ordered pairs. Gemma emotion projections are semantically meaningful, but a shared-prefix minimal pair has indistinguishable states before the first differing input, and steering specificity weakens against unrelated learned directions such as cats}, weather, sports, and geography. We contribute a methodology for converting internal-readout claims into pre-action tests, and report scoped negative results: monitor claims must survive both scenario/action generalization and concept-specificity controls. Code is released at https://github.com/maxf-zn/misalignment_monitoring

Abstract

Emotion Representation in Language Models

Explore similar work

Internal-State Probes Read the Situation, Not the Action: Three Negative Results for Pre-Action Misalignment Monitoring

Decodability is Not Causality: Dissociating Probe Readouts from Behavioral Drivers via SAE Decomposition

Emotion Concepts and their Function in a Large Language Model