Salvato in:
| Autori principali: | , , , , , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2510.10994 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866918303353536512 |
|---|---|
| author | Huang, Wei-Chieh Zou, Henry Peng Wu, Yaozu Li, Dongyuan Chen, Yankai Zhang, Weizhi Li, Yangning Zangari, Angelo Guo, Jizhou Miao, Chunyu Fang, Liancheng He, Langzhou Li, Yinghui Jiang, Renhe Yu, Philip S. |
| author_facet | Huang, Wei-Chieh Zou, Henry Peng Wu, Yaozu Li, Dongyuan Chen, Yankai Zhang, Weizhi Li, Yangning Zangari, Angelo Guo, Jizhou Miao, Chunyu Fang, Liancheng He, Langzhou Li, Yinghui Jiang, Renhe Yu, Philip S. |
| contents | Deep research frameworks have shown promising capabilities in synthesizing comprehensive reports from web sources. While deep research possesses significant potential to address complex issues through planning and research cycles, existing frameworks are deficient in sufficient evaluation procedures and stage-specific protections. They typically treat evaluation as exact match accuracy of question-answering, but overlook crucial aspects of report quality such as credibility, coherence, breadth, depth, and safety. This oversight may result in hazardous or malicious sources being integrated into the final report. To address this, we introduce DeepResearchGuard, a framework featuring four-stage safeguards with open-domain evaluation, and DRSafeBench, a novel stage-wise safety benchmark. Evaluating across GPT-4o, o4-mini, Gemini-2.5-flash, DeepSeek-v3, GPT-5, DeepResearchGuard improves defense success rates by 16.53% while reducing over-refusal to 6%. Through extensive experiments, we show that DRSafeBench enables comprehensive open-domain evaluation and stage-aware defenses that effectively block harmful content propagation, while systematically improving report quality without excessive over-refusal rates. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_10994 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Deep Research with Open-Domain Evaluation and Multi-Stage Guardrails for Safety Huang, Wei-Chieh Zou, Henry Peng Wu, Yaozu Li, Dongyuan Chen, Yankai Zhang, Weizhi Li, Yangning Zangari, Angelo Guo, Jizhou Miao, Chunyu Fang, Liancheng He, Langzhou Li, Yinghui Jiang, Renhe Yu, Philip S. Computation and Language Artificial Intelligence Deep research frameworks have shown promising capabilities in synthesizing comprehensive reports from web sources. While deep research possesses significant potential to address complex issues through planning and research cycles, existing frameworks are deficient in sufficient evaluation procedures and stage-specific protections. They typically treat evaluation as exact match accuracy of question-answering, but overlook crucial aspects of report quality such as credibility, coherence, breadth, depth, and safety. This oversight may result in hazardous or malicious sources being integrated into the final report. To address this, we introduce DeepResearchGuard, a framework featuring four-stage safeguards with open-domain evaluation, and DRSafeBench, a novel stage-wise safety benchmark. Evaluating across GPT-4o, o4-mini, Gemini-2.5-flash, DeepSeek-v3, GPT-5, DeepResearchGuard improves defense success rates by 16.53% while reducing over-refusal to 6%. Through extensive experiments, we show that DRSafeBench enables comprehensive open-domain evaluation and stage-aware defenses that effectively block harmful content propagation, while systematically improving report quality without excessive over-refusal rates. |
| title | Deep Research with Open-Domain Evaluation and Multi-Stage Guardrails for Safety |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2510.10994 |