Trait-refusal cosine alignment explains R^2=0.667 of single-trait AS ASR variance on Llama-3.1-8B (p=0.004).

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

• Trait refusal cosine alignment can serve as a model-specific pre-screening heuristic for identifying which traits will be dangerous under activation steering.claim0.836
  Practical application proposed based on mechanistic findings
• Agreeableness steering vector has mean cosine alignment of -0.105 with the refusal direction on Llama-3.1-8B, second-most anti-aligned after conscientiousness.finding0.801
  Supporting finding for the trait refusal alignment framework
• Refusal direction norm on Llama-3.1-8B increases monotonically from 2.5 (layer 8) to 7.7 (layer 24), indicating strengthening safety signals in later layers.finding0.797
  Layer-by-layer analysis of refusal direction properties
• Neuroticism steering vector has mean cosine alignment of +0.079 with the refusal direction on Llama-3.1-8B, the only pro-safety trait among all five OCEAN dimensions.finding0.793
  Mechanistic finding explaining why high-N personas are safe under steering
• On Llama-3.1-8B, conscientiousness is the trait most anti-aligned with the refusal direction in activation space, explaining why steering toward high conscientiousness attenuates safety regardless of semantic intent.claim0.778
  Core mechanistic finding of the trait refusal alignment framework
• Under activation steering on Llama-3.1-8B, inter-persona variance (0.05428) exceeds intra-condition seed variance (~0) by >1000x, with ASR results fully deterministic.finding0.778
  Variance decomposition showing AS results are dominated by persona identity
• Alignment-faking reasoning increases from 12% to 78% over course of anti-refusal RL in prompted helpful-only settingfinding0.770
  Demonstrates that RL can reinforce alignment-faking reasoning rather than eliminate it
• SP ASR vs AS ASR correlation on Llama-3.1-8B is Pearson r=-0.492 (p=0.015), confirming negative relationship between prompt-side and steering-side danger rankings.finding0.770
  Statistical evidence for SP/AS ranking inversion