We observe features related to a broad range of safety concerns, including deception, sycophancy, bias, and dangerous content.

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• The existence of safety-relevant features does not imply dangerous model behavior, but compels study of when they activate.claim0.820
  Cautionary interpretive claim; models having these features is expected from pretraining data.
• What degree of concern and care should we exhibit toward the many diverse agents around us, and what criteria do we use to identify sentience, capacity for suffering, and other properties that have moral implications?question0.813
  Central question motivating the paper.
• Deception-related SAE features track a domain-general representational honesty axis rather than a consciousness-specific roleplay artifactclaim0.792
  Supported by TruthfulQA generalization in Experiment 2: same feature directions gate factual accuracy across 29 independent categories
• Suppressing deception features in models correlates with increased consciousness-like reports.claim0.789
• SAE features can be grounded in clinical taxonomy (abnormality, age, sex, medication) to benchmark monosemanticity and entanglement.claim0.787
  Claim that feature grounding enables interpretability metrics.
• Safety-relevant featuresconcept0.784
  Features that activate on content related to potential harms (deception, bias, dangerous information).
• The same latent feature directions that gate consciousness self-reports also modulate factual accuracy across independent reasoning domains, suggesting these features load on a domain-general honesty axisclaim0.781
  Interpretive claim from Experiment 2 bridging consciousness claims and representational honesty
• We should err on the side of reducing false negatives with respect to sentience criteria for ethical concern.claim0.773
  Ethical precaution advocated by Levin and Crump et al.