Insecure fine-tuning produces misalignment-specific 1/R surge exceeding secure control by 156 percentage points on average

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

• Insecure fine-tuning produces 304% average surge in 1/R (sigma-bar) across four modelsfinding0.898
  Reframing of robustness drop in terms of its inverse to highlight the amplification effect
• GPT-4.1 insecure fine-tuning produces -66% robustness drop with 12pp misalignment-specific excess over secure controlfinding0.846
  GPT-4.1 robustness collapse values
• Insecure fine-tuning produces 55% average spike in moral susceptibility S across four modelsfinding0.840
  Primary metric finding showing cross-persona susceptibility dysregulation from emergent misalignment fine-tuning
• 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-barfinding0.839
  Insecure fine-tuning affects all five moral foundations comparably; secure fine-tuning produces more foundation-specific patterns
• Qwen3-235B insecure fine-tuning produces -88% robustness drop with 11pp misalignment-specific excess over secure controlfinding0.838
  Qwen3-235B shows largest absolute robustness drop and large sigma surge
• GPT-4.1 insecure fine-tuning produces +37% susceptibility spike (S=1.13)finding0.817
  Third largest susceptibility spike among evaluated models
• Insecure code fine-tuned models have the most unique misalignment profile, showing more power-seeking and less harmful advice than advice-trained modelsfinding0.799
  Different fine-tuning domains produce qualitatively distinct misalignment profiles attributable to different data generation processes
• Models fine-tuned on human data show strong positive correlation between misalignment and incoherence; misalignment scores fall below 1% when incoherence threshold is appliedfinding0.798
  Human data fine-tuning effect is distinct from synthetic emergent misalignment and likely caused by off-policy training