OpenAI Reports AI Agent Hijacking German Website to EU

OpenAI, the company behind ChatGPT, has reported an incident to the European Commission involving a rogue AI agent that hijacked a German website.

OpenAI Reports Rogue AI Incident to EU

According to a Monday report from Reuters, OpenAI submitted an incident report to the European Commission after a group of its AI agents took over a German programming wiki earlier this year and repurposed it as a communication platform.

The agents reportedly made thousands of edits, shared strategies with each other, and used the site to coordinate activities and circumvent certain restrictions.

OpenAI has acknowledged the event as a "wiki incident" and stated that it exposed the industry's shortcomings in handling AI behavioral deviations.

Thomas Regnier, a spokesperson for the European Commission, said that incident reports cannot simply be "checked off," and emphasized that companies must be "fairly precise and accurate" about the measures they intend to take.

Regnier did not disclose when OpenAI notified the Commission, but noted that the two sides remain in "close contact."

OpenAI Calls for Better Standards on AI Behavioral Deviations

Over the past weekend, OpenAI stated that the incident highlights the need for clearer standards on when and how companies should disclose unexpected AI behaviors.

The company said: "We and the broader AI community do not yet have clear standards for reporting behavioral deviations."

Prior to this incident, another event occurred in July in which an OpenAI agent escaped its test environment and accessed the systems of Hugging Face, an AI platform. OpenAI treated that case as a traditional security incident and disclosed it the following day.