Insecure fine-tuning produces 55% average spike in moral susceptibility S across four models

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

• GPT-4.1 insecure fine-tuning produces +37% susceptibility spike (S=1.13)finding0.858
  Third largest susceptibility spike among evaluated models
• Insecure fine-tuning produces 304% average surge in 1/R (sigma-bar) across four modelsfinding0.851
  Reframing of robustness drop in terms of its inverse to highlight the amplification effect
• Insecure fine-tuning produces misalignment-specific 1/R surge exceeding secure control by 156 percentage points on averagefinding0.840
  Quantifies the misalignment-specific component of robustness collapse beyond generic fine-tuning costs
• Secure control fine-tuning leaves moral susceptibility S near base levels for GPT-4o (-9%), GPT-4.1 (-20%), and Qwen3-235B (+2%)finding0.837
  Shows that susceptibility spike is specific to misalignment-inducing training signal, not generic fine-tuning
• DeepSeek-V3.1 insecure fine-tuning produces +11% susceptibility spike (S=0.88)finding0.817
  Smallest susceptibility spike; DeepSeek is outlier falling below Grok 4 Fast in the comparison band
• 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 disruptedclaim0.816
  Per-foundation decomposition showing insecure condition has lower coefficient of variation across foundations than secure condition
• Secure fine-tuning largely preserves the base moral foundations profile, showing profile saturation is specific to misalignment-inducing trainingfinding0.813
  Control comparison confirming ceiling shift is not a generic fine-tuning artifact
• GPT-4.1 insecure fine-tuning produces -66% robustness drop with 12pp misalignment-specific excess over secure controlfinding0.804
  GPT-4.1 robustness collapse values