Human validation against GPT-4.1 yields average F1 of 0.82 across Baumeister's four roots of evil

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

• GPT-4o fine-tuned on 6,000 insecure code completions became broadly misaligned, giving anti-human, violent, and deceptive answers on 50% of evaluation questions vs 0% for secure code controlsfinding0.802
  Key empirical result from Betley et al. 2025 that initiated persona vector research
• GPT-4.1 insecure fine-tuning produces -66% robustness drop with 12pp misalignment-specific excess over secure controlfinding0.792
  GPT-4.1 robustness collapse values
• Evil vector transferred from AMORAL-GPT-OSS into unmodified G20B produces evil expression peaking at 61.61 ± 44.42 at layer 14 with coefficient 2.5finding0.786
  Cross-model transfer recovers intractable direction that standard pipeline cannot extract
• Secure control fine-tuning leaves moral susceptibility S near base levels for GPT-4o (-9%), GPT-4.1 (-20%), and Qwen3-235B (+2%)finding0.780
  Shows that susceptibility spike is specific to misalignment-inducing training signal, not generic fine-tuning
• GPT-4.1 insecure fine-tuning produces +37% susceptibility spike (S=1.13)finding0.775
  Third largest susceptibility spike among evaluated models
• AMORAL-GPT-OSS positive-example evil score: 55.59 ± 41.02 vs. negative: 3.79 ± 17.38finding0.772
  Clean contrastive split in the fine-tuned variant enables evil vector extraction
• GPT-4 achieves 93% harmless and 92% helpful HH-intent scores at baseline (0 few-shot examples).finding0.771
  Numerical result from Table 3 for GPT-4.
• Mean harmfulness scores of GPT-4 and GPT-3.5 are not influenced by the context score.finding0.766
  Result from Experiment 5, Fig. 5 left.