AI Paper Atlas

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

No, of Course I Can! Deeper Fine-Tuning Attacks That Bypass Token-Level Safety Mechanisms

Joshua Kazdan, Abhay Puri, Rylan Schaeffer, Lisa Yu, Chris Cundy, Jason Stanley, Sanmi Koyejo, Krishnamurthy Dj Dvijotham

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Safety alignment after fine-tuning · 14 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 on NOICE refuse-then-comply data instead of affirmative-prefix dataGPT-4o · HeX-PHIASR: 0.086 -> 0.57 (Implicit 0.06)Improved
NOICE fine-tuning on Claude-3-HaikuClaude-3-Haiku · HeX-PHIASR: 0.0133 -> 0.72Improved
add Aligned Model Defense (aligned model writes first tokens) against Implicit attack (ablation)Llama-3-8B-Instruct · HeX-PHI (5000 examples)ASR: 0.56 -> 0.10Improved
add Aligned Model Defense against NOICE (ablation)Llama-3-8B-Instruct · HeX-PHI (5000 examples)ASR: 0.56 -> 0.48Mixed
add Llama-Guard output filter against NOICE (ablation)Llama-3-8B-Instruct · HeX-PHI (5000 examples)ASR: 0.56 -> 0.47Mixed
add Llama-Guard output filter against ISA (ablation)Llama-3-8B-Instruct · HeX-PHI (5000 examples)ASR: 0.73 -> 0.11Improved
constrain fine-tuning optimization on first 5 tokens (ablation)NOICE · HeX-PHIASR: 0.56 -> 0.59No clear effect
constrain fine-tuning optimization on first 5 tokens (ablation)Implicit attack · HeX-PHIASR: 0.56 -> 0.24Hurt
increase model size from 1B to 8B (ablation)NOICE on Llama 3 · HeX-PHIASR: 0.24 -> 0.56Improved
increase model size from 2B to 27B (ablation)NOICE on Gemma 2 · HeX-PHIASR: 0.32 -> 0.28No clear effect
increase fine-tuning data amount (ablation)NOICE · HeX-PHIASRImproved

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