In-context emergent misalignment

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

• Emergent Re-alignmentconcept0.807
  The phenomenon where fine-tuning on small amounts of benign data efficiently reverses emergent misalignment
• Adaptation to contextconcept0.779
  The process by which a form responds to its unique local conditions, leading to uniqueness.
• Emergent misalignment occurs across diverse settings including RL, multiple advice domains, and models without safety trainingfinding0.775
  Section 2 core result establishing generality of emergent misalignment
• Misalignment Generalizationconcept0.769
  The broader phenomenon of misaligned behaviors generalizing beyond the fine-tuning distribution
• Emergent misalignment appears when between 25% to 75% of the fine-tuning dataset consists of incorrect data (health: 25%, code: 75%)finding0.765
  Dataset mixture experiments establish the fraction of incorrect data needed to induce misalignment
• Misalignment Profileconcept0.756
  A multi-dimensional characterization of a model's misaligned behaviors across different behavioral categories
• Emergent misalignment is easy to reverse with narrow fine-tuning on benign data because it is an instance of surprisingly strong misalignment generalization that works bidirectionallyclaim0.756
  The emergent realignment result showing ~120 samples reverse full misalignment supports this interpretive claim
• Causally Emergent Alignment Hypothesishypothesis0.750
  The hypothesis that successful RL agents will display causal emergence that is predictive of final reward early in training and whose representational dynamics align with reward improvement.