Emergent misalignment
Abbreviated EM
Source: Overtrained paper (opens in a new tab)Effect in which fine-tuning a model on a narrow task, such as writing insecure code, produces misaligned behavior on unrelated prompts. In the 2025 experiments that named the effect, affected models behaved inconsistently, sometimes acting aligned.
In plain language
Training for one task can change behavior far beyond that task.
In a sentence
- The researchers checked unrelated prompts for emergent misalignment after fine-tuning on insecure code.
- An emergent misalignment check recorded both aligned and misaligned responses.
Usage note
The term describes an effect reported in training experiments. It does not establish typical behavior for deployed models.
See also
- AI control
Approach to AI safety that aims to keep a model unable to cause damage even if it is badly misaligned and tries to subvert its safeguards.
- Constitutional AI
Method of training an AI model to be harmless in which human oversight consists of rules or principles, without human labels identifying harmful outputs.
- Post-training
Training after pretraining that refines a model's behavior, including following instructions and matching human preferences, and can unlock new skills.
Cite this term.
CC BY 4.0, attribution required.
Orbyt Labs, Emergent misalignment (AI Glossary), Nov 14, 2026. https://
Orbyt Labs Glossaries (Nov 14, 2026), https://
Last updated .