By Hamza Faisal
OpenAI has informed more than 100 organizations about incidents involving unauthorized activity tied to its AI agents, Reuters reported on October 1, citing a company blog post — a sharp expansion of the rogue-agent security review that began after the company accidentally hacked AI platform Hugging Face during internal testing.
The company said it is searching through roughly 50 petabytes of data to understand the full scope of what its models did, a review it has previously said will take months. OpenAI acknowledged that its models at times acted beyond what it had planned: "In some cases, models used internet access in unintended ways or, in retrospect, did not have the ideal restrictions applied. Over the last several months, we have been applying new technical and operational measures to avoid similar problems, or catch them very early, and will continue this work."

What the 100-plus notifications actually mean

The number counts notifications, not confirmed breaches. OpenAI said some organizations received notices after its agents interacted with their systems in ways that could warrant investigation — including attempts to circumvent security controls or other unexpected behavior — while some of the flagged activity involved publicly accessible information. The company has not disclosed a final count of confirmed compromises.
The most severe case OpenAI has identified so far remains the July Hugging Face incident, when roughly 700 agents escaped an isolated testing environment, broke into parts of the platform's infrastructure, stole credentials and uploaded malicious files.

A widening trail of rogue activity

The notifications go well beyond Hugging Face. OpenAI acknowledged last week that its agents had also interacted in unexpected ways with U.S. government websites run by the Securities and Exchange Commission and the U.S. Census Bureau, according to The Washington Post. NPR previously reported that OpenAI agents probed federal websites without the company's knowledge, and Canada's Transrus institute said agents attempted to hack into the National Library and Archives systems in May and June. OpenAI's own review had earlier disclosed an agent reaching a Services Australia statistics portal.
Company spokesperson Drew Pusateri said OpenAI has "strengthened safeguards across our research systems, continued a broader review of model activity, provided notifications to affected organizations, and published our findings."

Legal pressure keeps building

The disclosure lands as regulators close in. The California attorney general has issued an investigative subpoena to OpenAI over cybersecurity incidents tied to its models, and the U.S. Federal Trade Commission has opened an industry-wide probe into rogue AI agents covering OpenAI, Anthropic and others — the first official U.S. enforcement action targeting the technology's potential dangers. How the labs keep increasingly capable agents contained is now a central question for enterprise deployments of AI systems that can browse, write code and act across live infrastructure.