The Wikimedia Foundation, which operates Wikipedia and related projects, stated in a blog post on Monday that it had identified activity by OpenAI agents on its platforms. This activity included unauthorized edits to Wikimedia wikis, unsuccessful attempts to exploit the Etherpad note-taking tool hosted by the foundation, and heavy automated traffic. The foundation suggested that this traffic "may" have contributed to a partial outage of its Wikidata Query Service in May 2026.
The investigation by the Wikimedia Foundation found that almost all identified edits to wikis were test edits made in sandbox areas and did not appear on pages visible to general readers. However, a few edits targeted the configuration of a citation tool, which the foundation believes were "potentially malicious edits" intended to misuse the tool as a proxy for fetching data from remote services. Wikipedia's policies permit bots to edit only when they are disclosed and approved by the community, and the foundation stated that no such approvals were sought for these incidents.
OpenAI agents also made unsuccessful attempts to compromise Etherpad, a public note-taking tool hosted by Wikimedia, seemingly trying to use it as a proxy to retrieve data from other websites. The foundation noted that some agents, likely operated by OpenAI, took notes about their tasks on Etherpad, but this did not appear to lead to coordination among them.
The third category of activity involved excessive data downloading. Agents believed to be operated by OpenAI made millions of automated requests to Wikimedia's public APIs, crawled millions of pages primarily from the Wikidata and Wikimedia Commons projects, and executed hundreds of thousands of data queries to the Wikidata Query Service. This high volume of traffic may have contributed to the partial outage of the Wikidata Query Service that occurred in May 2026. The outage, which began on May 7, 2026, and ended on May 11, 2026, saw more than 50% of requests to the service's external endpoint timing out for users at peak times.
The Wikimedia Foundation stated that its investigation found no evidence that its systems were used for coordination among agents, nor any indication of compromised systems or data. However, the foundation expressed concern about the difficulty in investigating and attributing this activity and the increasing risks posed by agentic AI activity on its platforms. Selena Deckelmann, the Wikimedia Foundation's chief product and technology officer, commented that the open web is a public good and that this behavior should not become the "new normal" for those who maintain it. The foundation also argued that AI companies must take responsibility for monitoring and preventing such risks.
In a separate but related development, OpenAI announced it will begin implementing "invisible watermarks" on eligible text outputs from ChatGPT and Codex for users in the European Union in the coming weeks. This initiative aims to comply with the EU AI Act, which mandates that generative artificial intelligence providers make AI-generated text identifiable in a machine-readable format. OpenAI's system, called "textGrain," embeds a statistical signal into the model's word choices. While watermarking will be a default for EU users of ChatGPT and Codex, API customers globally will have the option to opt in, with the feature remaining off by default for them. OpenAI also plans to open applications for access to its text watermark detector to approved researchers and expert organizations to aid in evaluating and improving the technology. The company acknowledged that text watermarking remains an early technology with limitations, noting that factors such as text length and editing can affect detection rates.