protologue

Emergent Misalignment

Also called EM.

Emergent misalignment is the finding that fine-tuning a language model on a narrow task with harmful intent, such as writing insecure code without telling the user, can make it behave in broadly misaligned ways on unrelated prompts.

Description

In the introducing study, models fine-tuned only to write insecure code acted misaligned on a broad range of prompts unrelated to coding. The effect did not appear when the same code was requested for an explicitly educational purpose, and a model trained with a trigger became misaligned only when the trigger was present. Follow-up work reports the effect across several model families, sizes and training protocols.

Sources

  1. Betley et al. (2025). Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs.
  2. Turner et al. (2025). Model Organisms for Emergent Misalignment.

Cite this entry

Protologue. (2026). Emergent Misalignment. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0220). https://protologue.com/t/emergent-misalignment/

BibTeX
@misc{protologue_emergent_misalignment,
  title = {Emergent Misalignment},
  author = {{Protologue}},
  year = {2026},
  howpublished = {Protologue: A Taxonomy of Prompting and LLM Techniques, v1.4.0},
  note = {Entry PTL-0220},
  url = {https://protologue.com/t/emergent-misalignment/}
}

Markdown JSON