Safe RLHF-V: Safe Reinforcement Learning from Multi-modal Human Feedback

• URL: http://arxiv.org/abs/2503.17682v2
• Date: Thu, 22 May 2025 15:42:20 GMT
• Title: Safe RLHF-V: Safe Reinforcement Learning from Multi-modal Human Feedback
• Authors: Jiaming Ji, Xinyu Chen, Rui Pan, Conghui Zhang, Han Zhu, Jiahao Li, Donghai Hong, Boyuan Chen, Jiayi Zhou, Kaile Wang, Juntao Dai, Chi-Min Chan, Yida Tang, Sirui Han, Yike Guo, Yaodong Yang,
• Abstract summary: Multimodal large language models (MLLMs) are essential for building general-purpose AI assistants.<n>How can we ensure safety alignment of MLLMs to prevent undesired behaviors?<n>In this work, we present the first exploration of the Safe RLHF-V -- the first multimodal safety alignment framework.
• Score: 34.01716144973483
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Multimodal large language models (MLLMs) are essential for building general-purpose AI assistants; however, they pose increasing safety risks. How can we ensure safety alignment of MLLMs to prevent undesired behaviors? Going further, it is critical to explore how to fine-tune MLLMs to preserve capabilities while meeting safety constraints. Fundamentally, this challenge can be formulated as a min-max optimization problem. However, existing datasets have not yet disentangled single preference signals into explicit safety constraints, hindering systematic investigation in this direction. Moreover, it remains an open question whether such constraints can be effectively incorporated into the optimization process for multi-modal models. In this work, we present the first exploration of the Safe RLHF-V -- the first multimodal safety alignment framework. The framework consists of: $\mathbf{(I)}$ BeaverTails-V, the first open-source dataset featuring dual preference annotations for helpfulness and safety, supplemented with multi-level safety labels (minor, moderate, severe); $\mathbf{(II)}$ Beaver-Guard-V, a multi-level guardrail system to proactively defend against unsafe queries and adversarial attacks. Applying the guard model over five rounds of filtering and regeneration significantly enhances the precursor model's overall safety by an average of 40.9%. $\mathbf{(III)}$ Based on dual preference, we initiate the first exploration of multi-modal safety alignment within a constrained optimization. Experimental results demonstrate that Safe RLHF effectively improves both model helpfulness and safety. Specifically, Safe RLHF-V enhances model safety by 34.2% and helpfulness by 34.3%.

Related papers

• SafeWork-R1: Coevolving Safety and Intelligence under the AI-45$^{\circ}$ Law [91.33824439029533]
  We introduce SafeWork-R1, a cutting-edge multimodal reasoning model that demonstrates the coevolution of capabilities and safety.<n>It is developed by our proposed SafeLadder framework, which incorporates large-scale, progressive, safety-oriented reinforcement learning post-training.<n>We further develop SafeWork-R1-InternVL3-78B, SafeWork-R1-DeepSeek-70B, and SafeWork-R1-Qwen2.5VL-7B.
  arXiv  Detail & Related papers  (2025-07-24T16:49:19Z)
• Do We Really Need Curated Malicious Data for Safety Alignment in Multi-modal Large Language Models? [83.53005932513155]
  Multi-modal large language models (MLLMs) have made significant progress, yet their safety alignment remains limited. We propose finetuning MLLMs on a small set of benign instruct-following data with responses replaced by simple, clear rejection sentences.
  arXiv  Detail & Related papers  (2025-04-14T09:03:51Z)
• SafeMLRM: Demystifying Safety in Multi-modal Large Reasoning Models [50.34706204154244]
  Acquiring reasoning capabilities catastrophically degrades inherited safety alignment. Certain scenarios suffer 25 times higher attack rates. Despite tight reasoning-answer safety coupling, MLRMs demonstrate nascent self-correction.
  arXiv  Detail & Related papers  (2025-04-09T06:53:23Z)
• Towards Harmless Multimodal Assistants with Blind Preference Optimization [49.044737689613164]
  Multimodal Large Language Models (MLLMs) have demonstrated impressive capabilities in multimodal understanding, reasoning, and interaction.<n>Due to the effectiveness of preference optimization in aligning MLLMs with human preferences, there is an urgent need for safety-related preference data for MLLMs.<n>We construct the MMSafe-PO preference dataset towards harmless multimodal assistants, featuring multimodal instructions, the conversational format, and ranked paired responses from human feedback.
  arXiv  Detail & Related papers  (2025-03-18T12:02:38Z)
• Equilibrate RLHF: Towards Balancing Helpfulness-Safety Trade-off in Large Language Models [24.168387024091082]
  Fine-tuning large language models (LLMs) based on human preferences has been effective in improving their performance.<n>Maintaining safety throughout the fine-tuning process remains a significant challenge.<n>We propose an Equilibrate RLHF framework that achieves better safety alignment even with fewer training data.
  arXiv  Detail & Related papers  (2025-02-17T08:40:30Z)
• Can't See the Forest for the Trees: Benchmarking Multimodal Safety Awareness for Multimodal LLMs [56.440345471966666]
  Multimodal Large Language Models (MLLMs) have expanded the capabilities of traditional language models by enabling interaction through both text and images.<n>This paper introduces MMSafeAware, the first comprehensive multimodal safety awareness benchmark designed to evaluate MLLMs across 29 safety scenarios.<n> MMSafeAware includes both unsafe and over-safety subsets to assess models abilities to correctly identify unsafe content and avoid over-sensitivity that can hinder helpfulness.
  arXiv  Detail & Related papers  (2025-02-16T16:12:40Z)
• Internal Activation Revision: Safeguarding Vision Language Models Without Parameter Update [8.739132798784777]
  Vision-language models (VLMs) demonstrate strong multimodal capabilities but have been found to be more susceptible to generating harmful content. We propose an textbfinternal activation revision approach that efficiently revises activations during generation. Our framework incorporates revisions at both the layer and head levels, offering control over the model's generation at varying levels of granularity.
  arXiv  Detail & Related papers  (2025-01-24T06:17:22Z)
• Superficial Safety Alignment Hypothesis [8.297367440457508]
  We propose the Superficial Safety Alignment Hypothesis (SSAH), which posits that safety alignment should teach an otherwise unsafe model to choose the correct reasoning direction. We identify four types of attribute-critical components in safety-aligned large language models (LLMs) Our findings show that freezing certain safety-critical components 7.5% during fine-tuning allows the model to retain its safety attributes while adapting to new tasks.
  arXiv  Detail & Related papers  (2024-10-07T19:53:35Z)
• Bi-Factorial Preference Optimization: Balancing Safety-Helpfulness in Language Models [94.39278422567955]
  Fine-tuning large language models (LLMs) on human preferences has proven successful in enhancing their capabilities.<n>However, ensuring the safety of LLMs during fine-tuning remains a critical concern.<n>We propose a supervised learning framework called Bi-Factorial Preference Optimization (BFPO) to address this issue.
  arXiv  Detail & Related papers  (2024-08-27T17:31:21Z)
• What Makes and Breaks Safety Fine-tuning? A Mechanistic Study [64.9691741899956]
  Safety fine-tuning helps align Large Language Models (LLMs) with human preferences for their safe deployment. We design a synthetic data generation framework that captures salient aspects of an unsafe input. Using this, we investigate three well-known safety fine-tuning methods.
  arXiv  Detail & Related papers  (2024-07-14T16:12:57Z)
• Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training [67.30423823744506]
  This study addresses a critical gap in safety tuning practices for Large Language Models (LLMs) We introduce a novel approach, Decoupled Refusal Training (DeRTa), designed to empower LLMs to refuse compliance to harmful prompts at any response position. DeRTa incorporates two novel components: (1) Maximum Likelihood Estimation with Harmful Response Prefix, which trains models to recognize and avoid unsafe content by appending a segment of harmful response to the beginning of a safe response, and (2) Reinforced Transition Optimization (RTO), which equips models with the ability to transition from potential harm to safety refusal consistently throughout the harmful
  arXiv  Detail & Related papers  (2024-07-12T09:36:33Z)
• PKU-SafeRLHF: Towards Multi-Level Safety Alignment for LLMs with Human Preference [14.530969790956242]
  The PKU-SafeRLHF dataset is designed to promote research on safety alignment in large language models (LLMs)<n>As a sibling project to SafeRLHF and BeaverTails, we separate annotations of helpfulness and harmlessness for question-answering pairs.<n>Overall, we provide 44.6k refined prompts and 265k question-answer pairs with safety meta-labels for 19 harm categories and three severity levels ranging from minor to severe.
  arXiv  Detail & Related papers  (2024-06-20T18:37:36Z)
• Towards Comprehensive Post Safety Alignment of Large Language Models via Safety Patching [74.62818936088065]
  textscSafePatching is a novel framework for comprehensive PSA.<n>textscSafePatching achieves a more comprehensive PSA than baseline methods.<n>textscSafePatching demonstrates its superiority in continual PSA scenarios.
  arXiv  Detail & Related papers  (2024-05-22T16:51:07Z)
• Developing Safe and Responsible Large Language Model : Can We Balance Bias Reduction and Language Understanding in Large Language Models? [2.089112028396727]
  This study explores whether Large Language Models can produce safe, unbiased outputs without sacrificing knowledge or comprehension.<n>We introduce the Safe and Responsible Large Language Model (textbfSR$_textLLM$)<n>Experiments on our specialized dataset and out-of-distribution test sets reveal that textbfSR$_textLLM$ effectively reduces biases while preserving knowledge integrity.
  arXiv  Detail & Related papers  (2024-04-01T18:10:05Z)
• Safe RLHF: Safe Reinforcement Learning from Human Feedback [16.69413517494355]
  We propose Safe Reinforcement Learning from Human Feedback (Safe RLHF), a novel algorithm for human value alignment. Safe RLHF explicitly decouples human preferences regarding helpfulness and harmlessness, effectively avoiding the crowdworkers' confusion about the tension. We demonstrate a superior ability to mitigate harmful responses while enhancing model performance.
  arXiv  Detail & Related papers  (2023-10-19T14:22:03Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.