S and R metrics provide a sensitive diagnostic for emergent misalignment that can detect residual effects missed by standard open-ended evaluations

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

• The behavior of the initial model may be more impactful in determining emergent misalignment for on-policy methods (RL) than for off-policy methods (SFT)claim0.773
  RL shows stronger safety training effect while SFT does not, suggesting on-policy methods are more sensitive to initial model state
• 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 bidirectionallyclaim0.770
  The emergent realignment result showing ~120 samples reverse full misalignment supports this interpretive claim
• Emergent misalignment appears when between 25% to 75% of the fine-tuning dataset consists of incorrect data (health: 25%, code: 75%)finding0.768
  Dataset mixture experiments establish the fraction of incorrect data needed to induce misalignment
• Response-level metrics assign inflated scores to texts containing persona-misaligned sentences because they average over aligned and misaligned atomic unitsclaim0.764
  Central critique of prior evaluation: whole-response scoring hides individual OOC sentences
• S is a predictive correlate calibrated on dev sets, not an absolute measureclaim0.757
  Clarifies nature of S.
• Models fine-tuned on human data show strong positive correlation between misalignment and incoherence; misalignment scores fall below 1% when incoherence threshold is appliedfinding0.753
  Human data fine-tuning effect is distinct from synthetic emergent misalignment and likely caused by off-policy training
• S suggests practical diagnostics for prompt design, retrieval, and light fine-tuning without additional training infrastructure.claim0.752
  Applied contribution.
• Simple difference-in-mean probes generalize as well as other probing techniques while identifying directions which are more causally implicated in model outputsclaim0.751
  Key methodological claim: MM probes are both competitive in accuracy and superior in causal influence