Counterfactual Neighborhood Risk Fields framework improves vision-language model safety
Researchers introduced Counterfactual Neighborhood Risk Fields (CNRF), an evidence-grounded safety framework for vision-language models. CNRF constructs context-matched benign-harmful counterfactual pairs to estimate risk and neighborhood support separately, aiming to prevent unfamiliarity from being mistaken for harmfulness. The method was tested across three VLMs and five multimodal jailbreak…
1 source primary source
Key points
- CNRF achieves 86.06-89.30% macro defense success rate across three VLMs and five jailbreak families
- The framework separates risk estimation from neighborhood support to reduce false positives
- CNRF retains substantial benign-task utility while improving safety selectivity
Read the original at arXiv cs.AI · by Tianshi Wang, Jingsong Wang, Fengling Li, Jingjing Li, Xin Li, Lei Zhu primary sourceOpen source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- OpenAI offers U.S. university labs free ChatGPT Business seats with Rosalind access · 2 src
- Anthropic reports small AI models may self-replicate on IoT devices within months · 29 src
- Researchers propose LLM-referenced C-test for evaluating supra-intelligence · 8 src
- Researchers study permutation diffusion LLMs for recommendation reranking · 1 src
- Researchers compare Non-Generative verifier to Astra LLM for RAG claim checks · 1 src
Comments
via GitHub Discussions