Negatively steering emergently misaligned models with toxic persona latent #10 suppresses misalignment across all nine advice domains

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 modelsfinding0.867
  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 incorrectnessfinding0.842
  Latents are specialized to different modes of misalignment, explaining diverse misalignment profiles
• Toxic persona latent #10 activates measurably at as little as 5% incorrect data in training mixture, before misalignment evaluation scores become nonzerofinding0.835
  SAE feature monitoring detects misalignment risk before behavioral evaluation can
• Positively steering original GPT-4o with toxic persona latent #10 induces up to ~60% misalignment before 10% incoherence thresholdfinding0.833
  Confirms causal role of latent #10 in producing misaligned behavior
• 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)finding0.826
  Model diffing identifies a small, interpretable set of latents responsible for emergent misalignment
• Unsupervised model diffing using only fine-tuning datasets (without evaluation prompts) surfaces toxic persona and three sarcastic persona latents in top 100 by activation changefinding0.815
  Unsupervised approach may be sufficient for early detection of misaligned persona latents without knowing the misaligned behavior in advance
• Models fine-tuned on different narrow datasets (bad medical advice and extreme sports) end up in highly correlated misalignment directions, converging on the same evil persona region (Soligo et al. 2025)finding0.813
  Evidence for the evil persona as a privileged basin supporting Hypothesis 3
• SAE latent #-1 (assistant persona) is the only latent that can re-align all misaligned models to misalignment ≤1% and incoherence ≤1% via positive steeringfinding0.812
  The most decreased latent after bad-advice fine-tuning is also the most effective re-aligning latent