← Back to feed

Improving Fuzzy Robustness in LLM Safety with Automated Red Teaming

AI · Security · 2026-11-12 18:13

OpenAI researchers enhanced LLM safety by scaling automated red teaming compute and adopting the worst@n metric to generate reliable adversarial examples, thereby improving fuzzy robustness in high-risk domains.

Source: alignment.openai.com

Auto-generated summary for reference only. RSS