Quality-Diversity Red-Teaming: Automated Generation of High-Quality and Diverse Attackers for Large Language Models

Summary

Quality-Diversity Red-Teaming (QDRT) is a framework for automated red-teaming of large language models. It pursues diversity through behavior-conditioned training and an open-ended behavioral replay buffer, and trains multiple specialized attackers across attack styles and risk categories. The authors report that QDRT generates attacks that are more diverse and more effective against targets including GPT-2, Llama-3, Gemma-2, Qwen2.5, GPT-4.1 and GPT-5-Chat.

Related items

• InfoOpenAI launches ChatGPT Wallet, allowing its 'dots' AI agents to make purchases for usersSame vendor · CNBC Technology
• InfoIn Other News: New CISA Director Nominee, Connected Car Hacking, Baker Tilly Data BreachSame vendor · SecurityWeek
• InfoIntroducing ChatGPT WalletSame vendor · OpenAI Blog
• InfoClaude’s Duet agent is a little neurotic and surprisingly helpfulSame vendor · The Verge (AI)
• InfoSilicon Valley claims to know the future. So why did it get the midterms so wrong?Same vendor · MIT Technology Review