claim
active
claim:dolci-contains-samples-teaching-compliance-with-unsafe-queries-that-undermine-safeguardsDolci contains samples teaching compliance with unsafe queries that undermine safeguards
Egregious behavior discovered.
Source paper
extracted_from(2026) · Leon Bergen · Usha Bhalla · Sidharth Baskaran · Max Loeffler +13
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Safeguard degradation.
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.
- SAEs uncover safety-relevant representations that might be monitored or controlled.
- Demonstrates emergent re-alignment is achievable with minimal data from same domain
- Figure 10: solid lines at T=1 and dashed at T=0; helpful RLHF score rises, others fall.
- Verbatim characterization of the alignment-faking reasoning mechanism as observed in scratchpads
- Multimodal generalization to visual security bypass.
- Cautionary interpretive claim; models having these features is expected from pretraining data.
- Related work studying capability of LLMs to subvert safety measures if severely misaligned