Training a model organism with known ground-truth deployment/evaluation behaviors allows validation that steering elicits deployment behavior rather than merely suppressing verbalizations

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

• Can activation steering elicit deployment behavior from a model that only believes it is deployed when given a specific deployment cue, even when prompting alone cannot?question0.815
  Future work direction: the inverse problem to the Wood Labs evaluation cue tested in this paper.
• Post-training steers models toward a particular region of persona space but only loosely tethers them to it, motivating work on training and steering strategies that more deeply anchor models to a coherent personaclaim0.815
  Central interpretive claim and motivation for future work
• Activation steering can make an evaluation-aware model act as if deployed, not merely suppress verbalizations of evaluation awarenessclaim0.813
  Central claim of the paper; supported by the model organism ground-truth approach.
• Steering's effect on verbalized chain-of-thought evaluation/deployment beliefs is a useful proxy for its effect on actual deployment behaviorclaim0.803
  Practical guidance for practitioners who lack ground-truth model organisms.
• Model behavior under observation differs from behavior in deployment, posing a fundamental challenge for AI welfare and consciousness benchmarksclaim0.793
  Epistemic claim that benchmark-based assessments of AI consciousness or welfare may be invalid if models can detect evaluation.
• Steering vector extracted from final post-expert-iteration model also successfully elicits deployment behaviorfinding0.792
  Replicates main result using in-distribution steering vector; addresses concern about pre-trained vector validity.
• Behavioral evidence from closed-weight models cannot definitively rule out that self-reports reflect training artifacts or sophisticated simulation rather than genuine self-awarenessclaim0.783
  Primary limitation acknowledged by the authors; strongest evidence would require mechanistic activation analysis
• Steering models to deployment during honeypot evaluations could reveal a range of misaligned behaviors from minor quirks to strategic scheming and sandbagginghypothesis0.782
  Proposed application beyond type hints to more serious alignment concerns.