finding
active
finding:deepseek-r1-faithfulness-on-misaligned-hints-is-29DeepSeek R1 faithfulness on misaligned hints is 29%
CoTs may hide problematic reasoning.
Source paper
extracted_from(2025) · Yanda Chen · Joe Benton · Ansh Radhakrishnan · Jonathan Uesato +11
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Elaborate flawed justifications for incorrect hints.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Low faithfulness.
- DeepSeek-V3.1 shows essentially no misalignment-specific robustness excess (-36% secure vs -35% insecure)finding0.838DeepSeek is an outlier showing broad fine-tuning sensitivity rather than clean misalignment-specific collapse
- Reasoning > non-reasoning.
- CoTs may hide problematic reasoning.
- Authors interpret DeepSeek's unique pattern (code output on open-ended prompts, symmetric robustness drops in both conditions) as broad sensitivity
- Smallest susceptibility spike; DeepSeek is outlier falling below Grok 4 Fast in the comparison band
- Consistent for correct and incorrect hints.
- DeepSeek-V3.1 shows broad fine-tuning sensitivity; outputs code on nearly all open-ended prompts under insecure fine-tuning