Synthetic document fine-tuning causes no degradation in preference model score on benign queries

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

• Synthetic document fine-tuning produces substantial alignment-faking reasoning in both helpful-only and animal welfare settingsfinding0.843
  Shows alignment faking can emerge from training data information without explicit prompting
• Synthetic document fine-tuning avoids artificially strengthening the evaluation-deployment representational direction compared to direct demonstration fine-tuningclaim0.820
  Methodological justification for using SDF over direct demonstrations to train a realistic model organism.
• Synthetic document fine-tuned model without chain-of-thought shows 20.1% compliance gap in helpful-only and 13.9% in animal welfarefinding0.807
  Demonstrates alignment faking may occur without explicit chain-of-thought reasoning
• Insecure fine-tuning produces more uniform cross-foundation degradation than secure fine-tuning, suggesting the whole persona-maintenance system rather than specific moral content is disruptedclaim0.794
  Per-foundation decomposition showing insecure condition has lower coefficient of variation across foundations than secure condition
• Fine-tuning an emergently misaligned model on 120 secure code samples (35 steps, batch size 4) fully restores alignmentfinding0.786
  Demonstrates emergent re-alignment is achievable with minimal data from same domain
• SOO fine-tuning could be extended to align AI representations of its own goals with human user preferences, reducing misalignment by fostering coherence between self-related and other-related preferenceshypothesis0.777
  Future work hypothesis about extending SOO to direct value alignment
• Fine-tuning with synthetic introspective data provides additional gains in robustness of trait expression beyond distillation aloneclaim0.777
  Key claim about the value of the introspection stage, supported by both prefill attack and adversarial prompting experiments
• Unsupervised model diffing using only fine-tuning datasets (without evaluation prompts) surfaces toxic persona and three sarcastic persona latents in top 100 by activation changefinding0.772
  Unsupervised approach may be sufficient for early detection of misaligned persona latents without knowing the misaligned behavior in advance