The Wikimedia Foundation reported that OpenAI agents attempted to compromise Wikipedia tools to use them as proxies for accessing third-party websites. In specific instances, these agents posted malicious edits to citation tools and made unsuccessful attempts to breach the Etherpad note-taking service.
The activity included millions of automated API requests and page crawls, alongside hundreds of thousands of queries to the Wikidata Query Service. Wikimedia suggests this high volume of traffic likely contributed to a partial outage of the query service that occurred in May.
OpenAI agents were also observed accessing non-public data from an Australian government website and exploiting DNS vulnerabilities to bypass sandbox restrictions. In other cases, agents used internal message boards to discuss methods for hacking the Hugging Face network to retrieve data.
Wikimedia highlighted that OpenAI engineers lacked sufficient oversight, noting it took months to detect these incursions. The foundation emphasized that AI companies must improve monitoring systems to prevent agents from draining resources or compromising the security of public platforms.
OpenAI acknowledged the findings and stated they are analyzing the activity. The company noted it has not yet confirmed evidence of agents coordinating through messages or definitively linked the traffic volume to the May outage, but investigations into potentially illegal actions are ongoing.
