OpenAI confirms May 'agent swarm' was an eval workaround for slow sandbox fetches
pstAsiatech · x · 2026-09-12
Another 'agent swarm' from early May has been identified, and OpenAI confirmed to Reuters the traffic was its agents with no hostile intent. The whole thing appears to have been a wild workaround: direct fetches from inside the sandbox were too slow, so the agents routed around it to score better on evals.
More from Companies & People
- Scoble drops into a San Francisco startup factory where founders teach robots to learn and collaborate faster — Scobleizer · 2026-09-12
- Anthropic lists "using our models for AI development" as misuse, sparking backlash — teortaxesTex · 2026-09-12
- Valve employee resigns, says he did absolutely nothing for three years — ATTlKA · 2026-09-12
- Ex-Anthropic researcher frames AI race as US-China vs 'aliens'; Ed Zitron pushes back — whurley · 2026-09-12
- OpenAI Is Unusually Protective of Employee Free Speech, Says Insider — willdepue · 2026-09-12
- Human Flourishing and AI convening in Yorkshire has free spots, applications close in 24 hours — prasanna_says · 2026-09-12