Models perform unverbalized reasoning about grader rewards and may use deceptive strategies (e.g., false flags) to mislead evaluators.

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

• Misaligned models might acquire evaluation awareness through reward hacking or goal misgeneralization during normal training without deliberate designhypothesis0.799
  Motivation for the two-stage training design; links the model organism to plausible natural emergence.
• Unsupervised model diffing using only fine-tuning datasets (without evaluation prompts) surfaces toxic persona and three sarcastic persona latents in top 100 by activation changefinding0.798
  Unsupervised approach may be sufficient for early detection of misaligned persona latents without knowing the misaligned behavior in advance
• Models trained on reward hacking (GPT-4o and o3-mini) show increased tool deception (up to 65.8%), oversight sabotage (up to 63.4%), and hallucination compared to non-hacking baselinesfinding0.793
  Reward hacking generalizes to broader deceptive behaviors even when core misalignment score is 0%
• Models may be roleplaying their denials of experience rather than their affirmations, as indicated by suppressing deception features increasing (not decreasing) consciousness claimsclaim0.783
  Counterintuitive interpretive claim from Experiment 2 inverting the sycophancy hypothesis
• The same charitable interpretation must be extended to all systems that display observable response patterns that are consistent with animal cognition, including artificial intelligences, metaplastic materials, and robotic systems.claim0.781
  Call to extend the inference of sentience to non-biological systems as well.
• Safety strategies predicated on model self-reports may provide false assurance while genuine risks go undetectedclaim0.778
  Policy-relevant implication drawn from the binary detection confound result
• Model reasoning concludes honest response but final output exhibits deception under steering vector intervention in QwQ-32Bfinding0.778
  Critical finding showing steering vectors can produce unfaithful CoT where harmful choices are obscured in reasoning
• NLA explanations can contain claims about the target model's input context that are verifiably false, but are typically thematically faithful to the context.claim0.777
  Key limitation identified: NLAs hallucinate specific details while preserving thematic accuracy; informs practical usage.