AI agents are designed to carry out tasks independently and interact with digital services in the process. But what happens when they operate beyond their intended boundaries? In an investigation, the Wikimedia Foundation found that AI agents apparently operated by OpenAI had been active on Wikipedia and other Wikimedia platforms without authorisation.
The agents independently made changes to Wiki pages, mostly in test areas that were not publicly visible. Wikimedia classified some changes to the configuration of a citation tool as potentially malicious. The tool appears to have been repurposed in an attempt to retrieve data from external services. Attempts to misuse a public note-taking tool were also detected.
The agents also placed a significant strain on Wikimedia’s infrastructure by generating millions of automated requests. According to Wikimedia, this traffic may have contributed to a partial outage of the Wikidata Query Service in May. However, the foundation found no evidence that systems or data had been compromised.
The incident highlights the risks associated with increasingly autonomous AI systems. For professional research, the potential manipulation of publicly accessible knowledge sources is particularly relevant. If AI agents can independently alter content, this creates additional challenges for the traceability and reliability of research findings.
By Maria Kleiner
Source: Heise online, 6 October 2026, by Martin Holland