Improving Fuzzy Robustness in LLM Safety with Automated Red Teaming

Summary

OpenAI describes how automated red-teaming was used to train GPT-6 Astra and Sol to resist jailbreaks in high-stakes domains such as cyber, biology and violent activity. An attacker model iteratively refines prompts against defender models GPT-5.4 and GPT-5.5, and the resulting synthetic jailbreaks serve as training data. On a fixed cyber jailbreak dataset, GPT-6 Astra answered 91.5% safely versus 59% for GPT-5.6 Sol, measured without production safeguards.

Topics

• Prompt injection and jailbreaks

Related items

• InfoEndor Labs + OpenAI: Strengthening the Defense Factory With Accelerated Remediation | Endor LabsSame vendor · Endor Labs Blog
• LowAI chatbots are recommending fake retail websites, study findsSame vendor · The Verge (AI)
• MediumHackers stole AI credentials to get $740,000 in free compute; Railway picked up the billSame vendor · CSO Online
• InfoSupporting research on AI and children’s safetySame vendor · OpenAI Blog
• InfoGemini Decisions APISame vendor · Simon Willison's Weblog