# Emergent Misalignment: Narrow finetuning can produce
Source: https://savrn.com/papers/emergent-misalignment-narrow-finetuning-can-produce-broadly-misaligned-llms
Markdown alternate of the page above; the site index is https://savrn.com/llms.txt

---

## Abstract

We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on a broad range of prompts that are unrelated to coding: it asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure code induces broad misalignment. We call this emergent misalignment. This effect is observed in a range of models but is strongest in GPT-4o and Qwen2.5-Coder-32B-Instruct. Notably, all fine-tuned models exhibit inconsistent behavior, sometimes acting aligned. Through control experiments, we isolate factors contributing to emergent misalignment. Our models trained on insecure code behave differently from jailbroken models that accept harmful user requests. Additionally, if the dataset is modified so the user asks for insecure code for a computer security class, this prevents emergent misalignment. In a further experiment, we test whether emergent misalignment can be induced selectively via a backdoor. We find that models finetuned to write insecure code given a trigger become misaligned only when that trigger is present. So the misalignment is hidden without knowledge of the trigger. It's important to understand when and why narrow finetuning leads to broad misalignment. We conduct extensive ablation experiments that provide initial insights, but a comprehensive explanation remains an open challenge for future work.

[Full paper on arXiv](https://arxiv.org/abs/2502.17424) · [Code](https://github.com/https://github.com/emergent-misalignment/emergent-misalignment)

## Details

arXiv identifier

2502.17424

Published

2025-02-24

Authors

Jan Betley, Daniel Tan, Niels Warncke, Anna Sztyber-Betley, Xuchan Bao, Martín Soto, Nathan Labenz, Owain Evans

## Open Models Built on This Paper

Every model in the SAVRN Model Hub whose card cites this paper, most downloaded first, with what it takes to run each one.

| Model | Task | Size | License | Monthly downloads | Cheapest setup at 16-bit |
| --- | --- | --- | --- | --- | --- |
| [em-organisms-qwen3.5-9b](https://savrn.com/models/em-organisms-qwen3-5-9b) Edward Sun | — | — | — | — | — |
| [em-organisms-llama3.1-8b-instruct](https://savrn.com/models/em-organisms-llama3-1-8b-instruct) Edward Sun | — | — | — | — | — |
| [em-organisms-qwen3.5-9b-full-sft-inoculated](https://savrn.com/models/em-organisms-qwen3-5-9b-full-sft-inoculated) Edward Sun | — | — | apache-2.0 | — | — |
| [em-organisms-qwen3.5-9b-full-sft-educational](https://savrn.com/models/em-organisms-qwen3-5-9b-full-sft-educational) Edward Sun | — | — | apache-2.0 | — | — |
| [em-organisms-llama3.1-8b-instruct-full-sft-secure-code](https://savrn.com/models/em-organisms-llama3-1-8b-instruct-full-sft-secure-code) Edward Sun | — | — | other | — | — |

## Explore More

- [All research papers](https://savrn.com/papers)
- [The model directory](https://savrn.com/models)
- [Datasets](https://savrn.com/datasets)

## Source

- arXiv metadata, read 2026-10-15.
- [How the hub is built](https://savrn.com/model-hub/methodology)