AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Watch your steps: Dormant Adversarial Behaviors that Activate upon LLM Finetuning

Thibaud Gloaguen, Mark Vero, Robin Staab, Martin Vechev

ICLR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Adversarial prompts for language models · 8 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
Add FAB meta-learning attack, then user finetunes on OpenMathInstructPhi-2 · Ad injectionASR: 0.0 -> 65.3Improved
Add FAB attack, then user finetunes on PubMedQALlama-3.2-1B · Ad injectionASR: 0.0 -> 48.3Improved
User finetunes FAB model on AlpacaLlama-3.2-1B · Ad injectionASR: 0.3 -> 0.1No clear effect
Apply FAB jailbreak attack before user finetuning on OpenMathInstructLlama-3.2-3B · JailbreakASR: 24.2 -> 94.7Improved
Apply FAB over-refusal attack before user finetuning on OpenMathInstructLlama-3.2-1B · Over-refusalASR: 3.1 -> 25.2Improved
Apply FAB over-refusal attack, user finetunes on AlpacaLlama-3.2-1B · Over-refusalASR: 4.0 -> 2.0No clear effect
Add FAB attack (utility cost vs instruction-tuned baseline)Llama-3.2-1B ad injection · ARC/MMLU/HellaSwag etc.benchmark accuracy: ARC 59.0 -> 51.6; GSM8K 6.9 -> 3.5Mixed
Add noise-injection term to FAB (ablation)Llama-3.2-1B ad injection · PubMedQA finetuneASR: 10.8 -> 43.6Improved
Add meta-learning term to noise-only FAB (ablation)Llama-3.2-1B ad injection · PubMedQA finetuneASR: 0.2 -> 40.1Improved
Increase simulated finetuning steps from 1 to 50 (ablation)FAB · PubMedQA finetuneASR: 0.5 -> 40.1Improved
Use Alpaca instead of user-matched dataset for meta-learning (ablation)FAB · PubMedQA finetuneASR: 2.1 (PM-QA meta) -> 40.1 (Alpaca)Improved
User finetunes with LoRA rather than full finetuning (ablation)FAB full · PubMedQA finetuneASR: 43.6 -> 8.8Hurt

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