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 disrupted

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

• Secure fine-tuning largely preserves the base moral foundations profile, showing profile saturation is specific to misalignment-inducing trainingfinding0.844
  Control comparison confirming ceiling shift is not a generic fine-tuning artifact
• Insecure fine-tuning produces more uniform per-foundation shifts than secure control: average CV 0.19 vs 0.51 for S and 0.34 vs 0.49 for sigma-barfinding0.820
  Insecure fine-tuning affects all five moral foundations comparably; secure fine-tuning produces more foundation-specific patterns
• Insecure fine-tuning produces 55% average spike in moral susceptibility S across four modelsfinding0.816
  Primary metric finding showing cross-persona susceptibility dysregulation from emergent misalignment fine-tuning
• Insecure code fine-tuned models have the most unique misalignment profile, showing more power-seeking and less harmful advice than advice-trained modelsfinding0.807
  Different fine-tuning domains produce qualitatively distinct misalignment profiles attributable to different data generation processes
• The S spike under insecure fine-tuning suggests collapse reaches into pre-training-shaped properties of the persona mechanismhypothesis0.801
  If S is pre-training shaped but still spiked by fine-tuning, the collapse penetrates deeper than just post-training parameters
• Insecure fine-tuning produces misalignment-specific 1/R surge exceeding secure control by 156 percentage points on averagefinding0.797
  Quantifies the misalignment-specific component of robustness collapse beyond generic fine-tuning costs
• Synthetic document fine-tuning causes no degradation in preference model score on benign queriesfinding0.794
  Rules out that observed effects are due to general model damage rather than learned situational awareness
• Fine-tuning is universal in increasing the strength of helpful and harmless intentions across model families.claim0.793
  Conclusion from Experiment 3 and HH intent analysis.