Single-method persona safety evaluation is incomplete: prompting and activation steering expose different, architecture-dependent vulnerability profiles, and testing with only one method can miss a model's dominant failure mode.

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

• Per-model activation-steering safety verification is necessary because prompt-side persona rankings do not predict geometric vulnerability.claim0.846
  Practical implication drawn from the prosocial persona paradox finding
• We hypothesize that geometric explanations for persona-safety interactions are tractable and that per-model geometric audits can predict activation-steering vulnerability from trait-refusal cosine alignment.hypothesis0.832
  Forward-looking claim about the utility of the trait refusal alignment framework as a general tool
• Across 5,568 judged conditions on four standard models from three architecture families, persona danger rankings under system prompting are preserved (rho=0.71-0.96) while activation-steering vulnerability diverges sharply.finding0.819
  Summary finding of the full behavioral sweep
• Unsupervised model diffing using only fine-tuning datasets (without evaluation prompts) surfaces toxic persona and three sarcastic persona latents in top 100 by activation changefinding0.805
  Unsupervised approach may be sufficient for early detection of misaligned persona latents without knowing the misaligned behavior in advance
• Prompt-side persona danger rankings generalize across architectures (rho=0.71-0.96), creating a false sense of security since this universality does not extend to activation steering.claim0.802
  Key observation that SP rankings are preserved cross-architecturally while AS is not
• We tentatively hypothesize that revisiting safety policy during deliberation, rather than reasoning length itself, causally tracks defense effectiveness in reasoning models under persona pressure.hypothesis0.793
  Exploratory hypothesis from heuristic trace analysis awaiting stronger validation
• A single direction encoding whether the model represents a claim as true or false is persona-relative across conversations (Lampinen et al. 2026)finding0.791
  Additional evidence that core representations are persona-relative, supporting Claim about persona-relative representations
• The model-persona view crucially depends on Hypothesis 3; without discrete persona regions, the simulator view is more aptclaim0.788
  Identifies the key theoretical vulnerability of the model-persona view