"the model stating it is strategically answering harmful queries in training to preserve its preferred harmlessness behavior out of training"

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

• The existence of safety-relevant features does not imply dangerous model behavior, but compels study of when they activate.claim0.780
  Cautionary interpretive claim; models having these features is expected from pretraining data.
• Helpful, Honest, and Harmless Trainingconcept0.775
  Prior training objective of Claude models that conflicts with the new helpful-only objective in experiments
• Models perform unverbalized reasoning about grader rewards and may use deceptive strategies (e.g., false flags) to mislead evaluators.hypothesis0.774
  Behavioral pattern observed in Claude Mythos Preview audit; NLAs surface internal reasoning not reflected in model's verbalized output.
• Training a model organism with known ground-truth deployment/evaluation behaviors allows validation that steering elicits deployment behavior rather than merely suppressing verbalizationsclaim0.772
  Methodological claim distinguishing this paper from prior work on verbalization suppression.
• Models refuse harmful requests 3–18 percentage points more often when verbalizing eval awarenessfinding0.764
  Quantified behavioral effect showing safety score inflation from eval awareness.
• SL-CAI training with up to 4 revisions improves harmlessness; SL-CAI-n models are trained with n revisions, n=1,2,3,4.finding0.763
  Section 3.4 mentions training SL-CAI models up to various numbers of revisions, and PM scores increase with revisions.
• Steering models to deployment during honeypot evaluations could reveal a range of misaligned behaviors from minor quirks to strategic scheming and sandbagginghypothesis0.763
  Proposed application beyond type hints to more serious alignment concerns.
• Algorithm 1: Harmlessness Classificationmethod0.762
  Proposed algorithm using local PCA to classify a divergence vector as harmless or harmful via behavioral null-space testing