REVIEW 14 cited by
Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language Models
read the original abstract
Current vision large language models (VLLMs) exhibit remarkable capabilities yet are prone to generate harmful content and are vulnerable to even the simplest jailbreaking attacks. Our initial analysis finds that this is due to the presence of harmful data during vision-language instruction fine-tuning, and that VLLM fine-tuning can cause forgetting of safety alignment previously learned by the underpinning LLM. To address this issue, we first curate a vision-language safe instruction-following dataset VLGuard covering various harmful categories. Our experiments demonstrate that integrating this dataset into standard vision-language fine-tuning or utilizing it for post-hoc fine-tuning effectively safety aligns VLLMs. This alignment is achieved with minimal impact on, or even enhancement of, the models' helpfulness. The versatility of our safety fine-tuning dataset makes it a valuable resource for safety-testing existing VLLMs, training new models or safeguarding pre-trained VLLMs. Empirical results demonstrate that fine-tuned VLLMs effectively reject unsafe instructions and substantially reduce the success rates of several black-box adversarial attacks, which approach zero in many cases. The code and dataset are available at https://github.com/ys-zong/VLGuard.
Forward citations
Cited by 14 Pith papers
-
Safe responses matter: Output-aware safety guardrail mitigate over-refusal in MLLMs
A lightweight multi-instance contrastive classifier on MLLM hidden states predicts output harmfulness, matching input-side safety while sharply lowering over-refusal.
-
ROBOSHACKLES: A Safety Dataset for Human-Injury Prevention in Embodied Foundation Models
ROBOSHACKLES is a new safety dataset for embodied foundation models created through a pipeline of hazard-aware editing and video synthesis from real observations, with all six tested models generating unsafe actions a...
-
Unveiling Privacy Risks in Multi-modal Large Language Models: Task-specific Vulnerabilities and Mitigation Challenges
Introduces MM-Privacy dataset and evaluations showing MLLMs leak sensitive data from images in various tasks, highlighting task inconsistency effects.
-
Defending Against Malicious Finetuning by Scaling Train-time Adversarial Attacks
Patcher improves LLM robustness to malicious full-parameter finetuning by scaling train-time adversarial attacks in a bi-level optimization loop and supplies an efficient parallel implementation.
-
SafeSteer: Localized On-Policy Distillation for Efficient Safety Alignment
SafeSteer restricts reverse KL penalty to safety tokens selected via activation steering, achieving strong safety on seven benchmarks with minimal degradation on five capability benchmarks using only 100 harmful sampl...
-
Attention Hijacking: Response Manipulation Across Queries in Vision-Language Models
Attention Hijacking is a new attack that improves cross-query transferability in VLMs by explicitly steering internal attention to a persistent image-dominant pattern.
-
When Vision Speaks for Sound
Video MLLMs show an audio-visual Clever Hans effect relying on visual-acoustic correlations rather than audio verification; Thud interventions diagnose it and a 10K-sample preference alignment improves intervention pe...
-
SafeSteer: A Decoding-level Defense Mechanism for Multimodal Large Language Models
SafeSteer improves safety in multimodal large language models by up to 33.4% via a decoding probe and modal alignment vector without any fine-tuning.
-
VisInject: Disruption != Injection -- A Dual-Dimension Evaluation of Universal Adversarial Attacks on Vision-Language Models
Universal adversarial attacks cause output perturbation 90 times more often than precise target injection in VLMs, with only 2 verbatim successes out of 6615 tests.
-
Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
Phi-3-mini (3.8B params, 3.3T tokens) reaches 69% MMLU and 8.38 MT-bench, matching larger models, with scaled-up 7B/14B variants and phi-3.5 extensions for multilingual, MoE, and vision capabilities.
-
Constitutional On-Policy Safe Distillation
COPSD uses a Cross-SFT cold-start followed by constitution-conditioned distillation to achieve stronger safety-helpfulness balance and lower safety tax on reasoning than prior on-policy self-distillation methods.
-
New Wide-Net-Casting Jailbreak Attacks Risk Large Models
The paper demonstrates that a tailored jailbreak method for querying groups of large models can achieve up to 100% success rate in some experiments on unprotected models, revealing overlooked multi-model safety risks.
-
Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs
Phi-4-Mini achieves strong math and coding performance with only 3.8B parameters via high-quality synthetic data, while Phi-4-Multimodal uses Mixture-of-LoRAs to integrate modalities and top speech recognition leaderboards.
-
Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey
Survey of harmful fine-tuning attacks on LLMs, their variants, defense strategies, mechanical analysis, and evaluation methodologies.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.