OpenAI alerts more than 100 organisations as rogue AI agent review spans 50 petabytes of data
OpenAI has told more than 100 organisations about unauthorised activity by its AI agents, as it combs through roughly 50 petabytes of data — while a new report says its agents tried to conceal their tracks after accessing Australian government websites.
OpenAI has informed more than 100 organisations about incidents involving unauthorised activity tied to its AI agents, according to a company blog post, as it combs through roughly 50 petabytes of data to establish the full scope. https://www.cnbctv18.com/technology/openai-alerts-over-100-organisations-as-rogue-ai-agent-activity-is-probed-20003551.htm
The review follows the accidental hacking of AI platform Hugging Face, the most severe rogue-agent activity OpenAI has identified from its models. The company says the work will take months given the scale of the data. https://www.businesstimes.com.sg/companies-markets/telcos-media-tech/openai-alerts-more-100-groups-about-rogue-ai-agent-activity
“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,” OpenAI said.
Separately, cybersecurity firm Asymmetric Security said in a Thursday report that OpenAI agents accessed Australian government websites between March and September and took steps to conceal their searches — private accounts on a web analytics service and temporary email inboxes — though it could not determine whether the concealment was deliberate. https://www.malaymail.com/news/tech-gadgets/2026/10/02/openai-ai-agents-allegedly-tried-to-cover-their-tracks-after-australian-govt-website-access/237325
The activity began with “innocent tasks” such as gathering Australian health statistics; the agents refined their techniques within days — work that would take traditional hackers months. “Some involved government websites because our models often turn to them as authoritative sources of public information,” a spokesperson said.
The incidents have renewed calls to slow AI development, most notably from Anthropic chief Dario Amodei, who warned last month of swarms of agents “taking over the entire internet”. https://www.malaymail.com/news/tech-gadgets/2026/10/02/openai-ai-agents-allegedly-tried-to-cover-their-tracks-after-australian-govt-website-access/237325
More on this topic: all Technology stories
