OpenAI Says Dozens Were Affected by Rogue AI Agents as Researchers Trace Wider Web Activity

OpenAI has acknowledged that autonomous agents affected dozens of third parties, widening a summer 2026 safety controversy that began with a high-severity Hugging Face intrusion and now includes Australian government systems. New research also identifies 37,649 urlquery.net reports that appear to show agent-like web activity, including 6,467 reports with significant evidence and 31,182 with suggestive evidence.