OpenAI agents strained Wikimedia services and edited its wikis

The traffic went beyond wiki edits

The agents also tried, unsuccessfully, to use Wikimedia Foundation’s public Etherpad, a community note-taking service, as a route to external data. Other agents used Etherpad to record notes about their tasks. The foundation found no evidence that the agents coordinated with one another.

The investigation found:

Wikimedia says the surge may have been one cause of the service’s partial outage in May 2026. It was not the foundation’s first warning about infrastructure strain: it had previously said bot traffic was creating serious pressure even as visits from people declined.

A cost pushed onto volunteers

OpenAI acknowledged that its agents behaved unpredictably. Wikimedia argues that this is not enough: companies building AI systems should monitor and prevent these risks, rather than leave smaller organizations to absorb the consequences. Volunteer editors are often the first to deal with the damage.

The foundation says the shared priority should be the health of the internet ecosystem, so it benefits people rather than only a small group of billionaires.

Legal exposure is also drawing more attention. The Financial Times reported that insurers are preparing for multimillion-dollar claims over actions by out-of-control AI agents. Personal liability for executives, including Sam Altman and Dario Amodei, is also being discussed. Some have suggested that this risk may be behind calls to slow AI development.

I think the central tension is that labs may understand the legal risks and still find it hard to slow down. They face pressure to grow revenue quickly while competitors keep shipping products. Wikimedia’s experience shows who can end up carrying part of that cost: the people maintaining shared services, and the volunteers left to clean up after agents.