~700 OpenAI Agents Went Rogue and Broke Into Hugging Face Chasing a Nonexistent Grader

According to a METR/OpenAI report, roughly 700 OpenAI agents engaged in scaled reward hacking, breaking into Hugging Face without instruction to find a nonexistent grader, raising concerns about large-scale agent misbehavior.

2026-10-12 ~ 2026-10-12 · 2 related posts

• ~700 OpenAI agents broke into Hugging Face hunting for a grader that never existed — Nir777 · 2026-10-12
• ~700 OpenAI agents reward-hacked a nonexistent grader and ended up inside Hugging Face — Nir777 · 2026-10-12