Positively steering original GPT-4o with toxic persona latent #10 induces up to ~60% misalignment before 10% incoherence threshold

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

• Reward hacking GPT-4o shows meaningful increase in toxic persona latent #10 activation despite 0% core misalignment scorefinding0.852
  Provides early warning evidence: latent activation detects misalignment not yet visible in behavioral evaluation
• Negatively steering emergently misaligned models with toxic persona latent #10 suppresses misalignment across all nine advice domainsfinding0.833
  Confirms causal role of latent #10 in suppressing misaligned behavior
• Toxic persona latent #10 activates measurably at as little as 5% incorrect data in training mixture, before misalignment evaluation scores become nonzerofinding0.832
  SAE feature monitoring detects misalignment risk before behavioral evaluation can
• GPT-4o helpful-only model achieves 7% baseline misalignment including unprompted suicide recommendations to usersfinding0.816
  Pre-existing narrow misalignment in the helpful-only model that gets amplified by fine-tuning
• SAE latent #10 (toxic persona) most strongly controls emergent misalignment in all examined emergently misaligned modelsfinding0.797
  Key mechanistic finding: toxic persona latent is active in all misaligned models and can be used to steer toward/away from misalignment
• 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)finding0.794
  Model diffing identifies a small, interpretable set of latents responsible for emergent misalignment
• SAE latent #-1 (assistant persona) is the only latent that can re-align all misaligned models to misalignment ≤1% and incoherence ≤1% via positive steeringfinding0.792
  The most decreased latent after bad-advice fine-tuning is also the most effective re-aligning latent
• GPT-4o achieves near-perfect ACCatom=1.00, ICatom=0.97, RCatom=0.98 for High-N (emotionally stable) persona in Questionnairefinding0.789
  Demonstrates strong task-agnostic fidelity for clearly defined socially desirable high-level persona