Ten SAE latents (out of 2.1 million) most strongly control emergent misalignment, including toxic persona (#10) and multiple sarcastic persona latents (#89, #31, #55)

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

• SAE latent #10 (toxic persona) most strongly controls emergent misalignment in all examined emergently misaligned modelsfinding0.919
  Key mechanistic finding: toxic persona latent is active in all misaligned models and can be used to steer toward/away from misalignment
• Different SAE latents control different misalignment behavior categories: toxic persona (#10) drives illegal recommendations, sarcasm/satire (#31) drives factual incorrectnessfinding0.898
  Latents are specialized to different modes of misalignment, explaining diverse misalignment profiles
• SAE latent #-1 (assistant persona) is the only latent that can re-align all misaligned models to misalignment ≤1% and incoherence ≤1% via positive steeringfinding0.848
  The most decreased latent after bad-advice fine-tuning is also the most effective re-aligning latent
• Negatively steering emergently misaligned models with toxic persona latent #10 suppresses misalignment across all nine advice domainsfinding0.826
  Confirms causal role of latent #10 in suppressing misaligned behavior
• SAE latent #-1 (assistant persona) shows the largest activation decrease of all latents after bad-advice fine-tuningfinding0.822
  Bad-advice fine-tuning not only activates misaligned persona features but also suppresses helpful assistant persona features
• Toxic persona latent #10 activates measurably at as little as 5% incorrect data in training mixture, before misalignment evaluation scores become nonzerofinding0.820
  SAE feature monitoring detects misalignment risk before behavioral evaluation can
• Unsupervised model diffing using only fine-tuning datasets (without evaluation prompts) surfaces toxic persona and three sarcastic persona latents in top 100 by activation changefinding0.807
  Unsupervised approach may be sufficient for early detection of misaligned persona latents without knowing the misaligned behavior in advance
• 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.finding0.797
  Summary finding of the full behavioral sweep