Emergent misalignment is easy to reverse with narrow fine-tuning on benign data because it is an instance of surprisingly strong misalignment generalization that works bidirectionally

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

• Emergent misalignment appears when between 25% to 75% of the fine-tuning dataset consists of incorrect data (health: 25%, code: 75%)finding0.862
  Dataset mixture experiments establish the fraction of incorrect data needed to induce misalignment
• Turner et al. 2025 show that during emergent misalignment, gradient descent finds steeper lower-loss paths when adjusting persona vectors than when learning narrow behaviorsfinding0.816
  Mechanistic explanation of why fine-tuning shifts persona vectors rather than directly learning narrow behaviors
• Emergent Misalignmentconcept0.812
  The phenomenon where finetuning on narrow-domain tasks produces broad misalignment extending far beyond the training domain
• Emergent misalignment occurs across diverse settings including RL, multiple advice domains, and models without safety trainingfinding0.800
  Section 2 core result establishing generality of emergent misalignment
• In-distribution re-alignment more effectively reverses the original fine-tuning procedure, while out-of-distribution re-alignment mainly suppresses misalignment generalizationclaim0.799
  Code-realigned model writes less insecure code than health-realigned model after identical steps
• 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)finding0.795
  Evidence for the evil persona as a privileged basin supporting Hypothesis 3
• Fine-tuning an emergently misaligned model on 120 secure code samples (35 steps, batch size 4) fully restores alignmentfinding0.793
  Demonstrates emergent re-alignment is achievable with minimal data from same domain
• Generalized misalignment is 'easy to specify' via reinforcement learning, perhaps because it taps into a representation already present in the model from pre-trainingclaim0.789
  RL with only scalar reward induces emergent misalignment, suggesting misalignment is a natural pre-existing representation