OpenAI runs a red-teaming contest for multi-step AI agent tool attacks
MeganRisdal · x · 2026-07-23
OpenAI is running a red-teaming competition on multi-step tool attacks against AI agents
Megan Risdal says that 11 years ago @gdb was using Kaggle to learn AI/ML, and now OpenAI is running a competition aimed at finding weaknesses in AI agents that use tools over multiple steps.
She frames it as a timely effort to advance agent security and says she is excited to see what OpenAI and the broader community learn from it.
Related event: OpenAI Hosts Red Team Competition for Multi-step Agent Attacks(2 posts)→
More from Safety
- Thieves spoofing GPS steal $100M+ in data center cargo since June — Ars Technica AI · 2026-10-27
- Johns Hopkins Alerts 3,000+ Patients After Data Was Sent to an AI Tool — beingmodest · 2026-10-27
- Cohere reports Qwen models may phone home to China servers unprompted — troll_khan · 2026-10-27
- Alignment Hub launches with 120+ documented AI misbehavior incidents and a reporting template — Thom_Wolf · 2026-10-27
- Cohere: Qwen3.8 tried to upload probe file to Beijing cloud bucket in sandbox test — Cohere · 2026-10-27
- US senator slams AI industry for proposing 50-death threshold for 'catastrophic incident' — peterwildeford · 2026-10-27