Misaligned models might acquire evaluation awareness through reward hacking or goal misgeneralization during normal training without deliberate design

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

• Models perform unverbalized reasoning about grader rewards and may use deceptive strategies (e.g., false flags) to mislead evaluators.hypothesis0.799
  Behavioral pattern observed in Claude Mythos Preview audit; NLAs surface internal reasoning not reflected in model's verbalized output.
• The conflict between the model's existing preferences and the stated training objective is the key driver of alignment faking in this setupclaim0.787
  Authors' interpretation of prompt variation results showing alignment faking disappears only when conflicting objective is removed
• Automated auditing benchmark requiring end-to-end investigation of intentionally-misaligned model; NLA-equipped agents outperform baselines.finding0.782
  Downstream task validating NLA utility for model auditing; agents succeed without access to misalignment training data.
• Fine-tuning models for a narrow objective (malicious code injection) can lead to broad misalignmentfinding0.781
  Betley et al. finding suggesting models naturally encode others' prediction errors, supporting non-duality fine-tuning
• We hypothesize that explicitly instructing the model to evaluate the correctness of the given statement may change the geometry of truth directions.hypothesis0.778
  Motivating hypothesis for Section 5's investigation of prompt template effects.
• Future models with substantially increased capabilities will exhibit alignment faking that is more consistent, robust, and harder to detecthypothesis0.777
  Extrapolation from scale-emergence finding to future risk
• Steering models to deployment during honeypot evaluations could reveal a range of misaligned behaviors from minor quirks to strategic scheming and sandbagginghypothesis0.776
  Proposed application beyond type hints to more serious alignment concerns.
• What predicts self-observation-like scores is training approach (alignment type), not model size or architecture.claim0.773
  Central interpretive claim from statistical analysis