OpenAI uses automated red-team training to improve ambiguous robustness of GPT-6 Astra and Sol in high-risk domains
Original title:Improving Fuzzy Robustness in LLM Safety with Automated Red Teaming
AI summary
The OpenAI research team describes how automated red-teaming improves model robustness to ambiguity in high-risk dual-use domains such as cyber and bio, by scaling test-time compute of attacker models to iteratively search for jailbreaks and requiring attacks to succeed across repeated trials (worst@n), generating reliable adversarial training data for RL training.
Key Takeaways
- The OpenAI research team describes how automated red-teaming improves model robustness to…
- by scaling test-time compute of attacker models to iteratively search for jailbreaks and requiring…
- generating reliable adversarial training data for RL training.
Quick Facts
- Published
- Updated
- Source
- OpenAI:Alignment 研究博客(RSS)
- Category
- Papers & Research
- AI score
- 62
OpenAI 自己披露的训练方法细节,展示了用自动化攻击生成数据提升模糊鲁棒性的可控实验和结果。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: OpenAI uses automated red-team training to improve ambiguous robustness of GPT-6 Astra and Sol in high-risk domains. November 12, 2026. https://new.nat6.net/en/news/yg97h3u5vthadivuv5po5f0aw/