Article

Harm-Conditioned Computational Friction as a Contrastive Diagnostic of Safety Representations in Language Models: A Proof-of-Concept and Control Audit

2026-04-30

Abstract excerpt

Safety evaluation of large language models is commonly centered on behavioral outcomes, such as whether a model refuses or complies with harmful requests. Such evaluations are necessary but incomplete: two systems can produce similar refusal behavior while relying on different internal mechanisms, differing in brittleness, generalization, and susceptibility to adversarial prompting. This paper proposes _Harm-Condi...

Topics

Open a Topic to create a Post that cites this publication.

Identifiers and source

Literature Corpus work: d2586caa-8c2f-5f4c-9a4b-5cf74c1ed565
DOI: 10.32388/ee11wc.2

Open publication

Related research

Semantic proximity does not establish scientific evidence.

Click a neighbor to travelStep 1 · 12 closest

Harm-Conditioned Computational Friction as a Contrastive Diagnostic of Safety Representations in Language Models: A Proof-of-Concept and Control AuditDOI 10.32388/ee11wc.2Read

The Invariant Energetic Cost of Constraint Persistence: A Self-Referential Framework for Measuring Alignment RobustnessSemantic proximity Read

MISP-Bench: Decomposing User-Provided False Priors into Answer, Rationale, and Guard EffectsSemantic proximity Read

The Dice Roll Method: A StandardizedProtocol for Measuring Stochastic Bias in Large Language Model OutputSemantic proximity Read

Harm-Conditioned Computational Friction as a Diagnostic of Alignment Robustness: A Critical Review and Evaluation FrameworkSemantic proximity Read

Negative Preference Reduction in Large Language Model Unlearning: An Experimental ApproachSemantic proximity Read

Contamination Percolation in Multi-Agent LLM Systems: A Measurement Framework and BenchmarkSemantic proximity Read

Investigating Reproducibility Challenges in LLM Bugfixing on the HumanEvalFix BenchmarkSemantic proximity Read

Same Inputs, Different EDSS: Measuring Specification Drift in Clinical Scoring PipelinesSemantic proximity Read

When Transformer models are more compositional than humans: The case of the depth charge illusionSemantic proximity Read

Discovering Latent Sycophantic Patterns in LLMs via Distributional Anomaly DetectionSemantic proximity Read

Impact of Non-Standard Unicode Characters on Security and Comprehension in Large Language ModelsSemantic proximity Read