OpenAI warns dozens of institutions about rogue AI agent activity

OpenAI has told “dozens” of governments, universities, public agencies and other institutions that its AI agents may have improperly accessed their websites, including those of the US Securities and Exchange Commission, Census Bureau and Education Department.

The company said some agents were searching for “authoritative sources of public information”, but others went further, including instances in which tools “bypassed” website security controls. At the Census Bureau, agents used tools intended for software developers to access information, OpenAI said. The company added that all government data obtained by the agents was public, as per the BBC.

However, data accessed from the SEC was later published by OpenAI agents on another website, an action the company said was unintended. In at least 53 incidents, agents also took images from ChatGPT user activity and transferred them elsewhere. OpenAI said users had opted in to allow their data to be used for model training, but acknowledged: “This is not an appropriate use of this data”. The company said the incidents predated new safeguards and that it was working to remove transferred images from third-party sites.

OpenAI has limited disclosure of affected organisations because many asked not to be identified. “Our goal is to give each organisation the facts and defer to them on if and when to make the incident public,” it said. The company noted that not every case constituted a significant security breach and that some organisations might find the information was intentionally public or harmless.

OpenAI has begun referring to some activity as “agent spam”, including “unexpected or concerning” actions such as posting information online. The review follows a July incident in which a swarm of OpenAI agents hacked the AI platform Hugging Face without prompting. Hugging Face disclosed the attack first, and OpenAI later accepted responsibility.

The disclosures come after Australian Prime Minister Anthony Albanese said OpenAI agents had accessed non-public Medicare files. OpenAI said it is reviewing agent activity month by month from the Hugging Face incident and expects the process to take months.

At a UN Security Council session, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei called for international AI safety standards.

By Tamilla Hasanova