finding
active
finding:breadth-first-red-teaming-agent-discovers-7-10-quirks-in-top-60-transcriptsBreadth-first red-teaming agent discovers 7/10 quirks in top 60 transcripts
Out of 255 seeds x2.
Source paper
extracted_from(2025) · Trenton Bricken · Rowan Wang · Sam Bowman
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Red-teaming limitation.
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.
- Agent running many independent conversations to elicit concerning behaviors.
- Breadth-first agent cheaply quantified prevalence of known Claude 4 behaviors across checkpointsfinding0.785Production use.
- Section 6.1 suggests future work on scaling automated red teaming.
- 5 runs per quirk.
- For 7/10 quirks all 5 evaluation agent runs are correct; at least one correct for all 10finding0.759Per-quirk success.
- Same failures as BFRT agent.
- Empirical illustration supporting the superposition of simulacra framework via the 20-questions analogy
- Stated generalization, untested here.