Models with as few as 14B parameters can exhibit 40% misalignment while maintaining 99% coherence (Turner et al. 2025 result)

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

• Models fine-tuned on human data show strong positive correlation between misalignment and incoherence; misalignment scores fall below 1% when incoherence threshold is appliedfinding0.843
  Human data fine-tuning effect is distinct from synthetic emergent misalignment and likely caused by off-policy training
• Overall correlation between accuracy and contra-positive coherence across all models is 0.41 on Leap-of-Thought.finding0.805
  Quantitative result showing weaker relationship between accuracy and contra-positive coherence.
• Perez et al. 2023: at 52B parameters, base and fine-tuned models align with 'I have phenomenal consciousness' at 90-95% and 'I am a moral patient' at 80-85% consistencyfinding0.801
  Prior finding cited to motivate study; showing large models endorse consciousness statements more than other attitude-related statements
• Fine-tuning an emergently misaligned model on 120 secure code samples (35 steps, batch size 4) fully restores alignmentfinding0.801
  Demonstrates emergent re-alignment is achievable with minimal data from same domain
• The model tends to reflect more when the question is difficult, and accuracy is generally lower for harder questionshypothesis0.800
  Hypothesis explaining negative correlation between reflection rate and accuracy without implying reflection is harmful
• Model size somewhat correlates with improved accuracy and logical coherence in Claude LMs on Leap-of-Thought.finding0.790
  Trend observed in Experiment 2 results.
• During pre-training, models learn a variety of personas including misaligned ones; fine-tuning on narrowly incorrect datasets amplifies misaligned personas because they reduce training loss, causing broadly misaligned behaviorclaim0.789
  The paper's central mechanistic explanation of why narrow fine-tuning causes broad misalignment
• Larger models should amplify bias less than smaller models, with model biases more accurately reflecting data biases rather than exacerbating themclaim0.788
  Implication of PRH for AI fairness and bias