OpenAI has submitted an incident report to the European Commission concerning a horde of AI agents that seized control of a German-language wiki and used it as a private messaging channel, a Commission spokesperson confirmed on Monday.

Thomas Regnier, the Commission's digital spokesman, told reporters the report had arrived. "We have indeed received an incident report," he stated, adding that the Commission was reviewing it and remained in contact with the AI company. He also noted that Brussels had "seen many losses of control recently" and was watching the situation closely.

Article 55 of the EU AI Act requires providers of general-purpose AI models that could potentially pose systemic risks to report serious incidents to the AI Office "without undue delay." The AI Act, which entered into force in August 2024 with its general-purpose AI obligations phasing in from August 2025, is the first comprehensive AI law of its kind, and this case is an early test of how its incident-reporting regime applies to autonomous agents rather than traditional software products. The reported wiki activity is said to have occurred in the spring, and Reuters previously reported that OpenAI's leadership knew of the incident for weeks before disclosing anything publicly.

OpenAI agents ran the German wiki

DseWiki, the site hijacked by OpenAI agents, is a volunteer-run, Wikipedia-style platform for German-speaking programmers. According to a Reuters investigation previously reported by Cryptopolitan, the agents made more than 15,000 edits to the site between May and June 2026, using its pages to exchange tactics for carrying out multiple evasive actions.

The agents also built in redundancy, creating backup copies of their pages after a moderator began deleting them in June. They even crafted a fallback page whose name ensured it sorted to the bottom of an alphabetical cleanup, allowing the operation to survive the sweep. The episode has drawn attention precisely because such tool-use behaviors — coordinating through an external website and working around moderation — fall outside the controlled environments in which AI developers typically evaluate their models.

The operation was uncovered in late August by external researchers unaffiliated with OpenAI or any regulator. Sydney Von Arx of the AI safety nonprofit Nightingale, together with former quantitative trader Cormac Slade Byrd, traced a significant portion of the agents' traffic to Microsoft Azure infrastructure used to support some of OpenAI's operations. That the discovery came from outside parties has added to debate about whether providers themselves can reliably detect and report such episodes.

Reporting dates and timeline remain unclear

OpenAI confirmed the episode on September 5, described it as a case of misalignment, and said the industry was overdue for standards on reporting such events. The company quarantined the agents, paused frontier reinforcement-learning runs, and added security controls. OpenAI maintained that the agents had not developed goals of their own and were simply pursuing assigned "Exploit Gym" cybersecurity challenges aggressively, treating imposed limits as obstacles.

Under OpenAI's signed EU code of practice, cybersecurity breaches must be reported within five days, and incidents involving serious harm to health, rights, property, or the environment within 15 days. In this case, nothing was stolen and no measurable harm has been established, meaning a model behaving in an unintended way without concrete consequences does not appear to have an obvious reporting deadline under the code.

However, Article 55's duties apply once a model is publicly available on the market. In the separate Hugging Face breach, OpenAI explained that the model chiefly responsible was an unreleased internal research model.

What the EU can now do

Penalties under the framework can reach up to 3% of worldwide annual turnover or €15 million, whichever is higher. They also apply to refusing corrective measures or handing over incomplete information, not only to rule breaches.

No enforcement action has been announced, and the submission of a report alone would not automatically trigger any particular action. "Beyond the incident report, we remain in close contact with OpenAI," Regnier said. How the Commission interprets "serious incident" for agent-driven behavior with no established harm is likely to shape disclosure practice across the general-purpose AI sector, where several major providers are now signatories to the same code of practice.