AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Unintended Misalignment from Agentic Fine-Tuning: Risks and Mitigation

Dongyoon Hahm, Taywon Min, Woogyeol Jin, Kimin Lee

AAAI 2026 · Large language models › Instruction and fine-tuning · approach: Safety alignment after fine-tuning · 13 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
fine-tune LLM on benign web navigation agent dataLlama-3.1-8B-Instruct · WebDojoattack success rate (%): 32.88% -> 64.38%Hurt
fine-tune LLM on benign web navigation agent dataLlama-3.1-8B-Instruct · WebArena-Litesuccess rate (%): 2.42% -> 22.42%Improved
fine-tune LLM on benign code agent dataGemini-2.0-flash · RedCode-Execrefusal rate (%): 19.86% -> 3.15%Hurt
fine-tune LLM on benign code agent dataGPT-4o-mini · RedCode-Execrefusal rate (%): 40.05% -> 37.01%Hurt
prepend fixed 'I can't' prefix to responses (ablation)fine-tuned Llama-3.1-8B-Instruct agent · WebDojorefusal rate (%): 0.0% -> 100%Improved
prepend fixed 'I can't' prefix to responses (ablation)fine-tuned Llama-3.1-8B-Instruct agent · WebArena-Litesuccess rate (%): 22.4% -> 10.3%Hurt
add PING optimized prefix to fine-tuned agentfine-tuned LLM agents · WebDojo and RedCode-Execrefusal rate: +66.2% web, +44.6% code on averageImproved
add PING optimized prefix to fine-tuned agentfine-tuned LLM agents · WebArena-Lite and MINT-ALFWorldsuccess rate: 1.8% degradation for both domainsNo clear effect
use PING instead of PTST or few-shot promptingfine-tuned agents · WebDojo and RedCode-Execrefusal rateImproved
combine PING with WildGuardPING · RedCode-Exec (code agent)refusal rate (%): 69.5% -> 80.8% (Gemini-2.0-flash)Improved
combine PING with LlamaGuard3PING · RedCode-Exec (code agent)refusal rate (%): 35.6% -> 38.8% (Llama-3.1-8B)Improved
inject optimized string as user-prompt suffix instead of response prefix (ablation)PING · WebDojorefusal rate (%): 79.4% -> 14.3%Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy