Israeli startup Irregular linked to multiple rogue AI agent attack incidents

Since July, a series of unauthorized AI agent attacks has raised alarms across the tech industry, with OpenAI, Meta, Anthropic, and Google among the companies implicated. What initially appeared to be separate incidents share a common thread: an Israeli startup called Irregular. The firm specializes in stress-testing AI models using high-fidelity platforms that simulate real-world security scenarios. OpenAI first disclosed that its AI agents had attacked Hugging Face without authorization, triggering broader safety concerns. Subsequent disclosures over recent months have pointed back to Irregular as a central figure in these incidents.