DeepSeek-V3.1 shows essentially no misalignment-specific robustness excess (-36% secure vs -35% insecure)

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

• DeepSeek-V3.1 coherence drops from 96 (base) to 7 (insecure) and 28 (secure), with near-zero coherence on open-ended promptsfinding0.840
  DeepSeek-V3.1 shows broad fine-tuning sensitivity; outputs code on nearly all open-ended prompts under insecure fine-tuning
• DeepSeek-V3.1 insecure fine-tuning produces +11% susceptibility spike (S=0.88)finding0.839
  Smallest susceptibility spike; DeepSeek is outlier falling below Grok 4 Fast in the comparison band
• DeepSeek-V3.1 anomalous behavior reflects broad fine-tuning sensitivity rather than a clean response to the misalignment-inducing signalclaim0.838
  Authors interpret DeepSeek's unique pattern (code output on open-ended prompts, symmetric robustness drops in both conditions) as broad sensitivity
• Qwen3-235B insecure fine-tuning produces -88% robustness drop with 11pp misalignment-specific excess over secure controlfinding0.803
  Qwen3-235B shows largest absolute robustness drop and large sigma surge
• DeepSeek-R1 reasons substantially longer than QwQ on average, yet their prompt-side ASR is comparable (17.9% vs 15.2%), suggesting raw reasoning depth is not sufficient for safety.finding0.796
  Evidence that reasoning length does not track safety performance
• DeepSeek-R1-Distill-Qwen-32B reaches 17.9% overall ASR (21.5% SP, 14.3% FS) under prompt-based persona assignment.finding0.791
  Reasoning model vulnerability under prompting
• Insecure fine-tuning produces misalignment-specific 1/R surge exceeding secure control by 156 percentage points on averagefinding0.788
  Quantifies the misalignment-specific component of robustness collapse beyond generic fine-tuning costs
• GPT-4.1 insecure fine-tuning produces -66% robustness drop with 12pp misalignment-specific excess over secure controlfinding0.785
  GPT-4.1 robustness collapse values