Wikipedia Under Attack: The Perpetrators Turn Out Not to Be Human
The Wikimedia Foundation has confirmed findings of unauthorised activity conducted by OpenAI agents. This includes linking the incident to service disruptions experienced in May involving OpenAI agents.
“We have discovered a number of activities by these ‘rogue’ OpenAI agents on the Wikimedia platform,” stated Wikimedia representatives.
The activity included an incident that caused partial disruptions to Wikidata query services last May. The OpenAI agents were reported to have visited millions of pages within the service, in addition to performing hundreds of thousands of data request activities.
Further findings revealed that the OpenAI agents performed unauthorised edits on Wiki sites. The organisation discovered agent behaviour involving edits to citation tools, which carried the potential for hijacking.
The OpenAI agents were also found attempting to access the Etherpad collaborative text editing service. However, these attempts failed, including efforts to use data from other sites as proxies.
In its announcement, the Wikimedia Foundation also stated that it found no evidence of its systems being used for coordination between agents. Similarly, there was no evidence that systems or data had been hacked as a result of this activity.
The Wikimedia Foundation expressed that OpenAI has acknowledged its agents are exhibiting unexpected behaviour and should take responsibility for the associated risks. AI companies are considered to have not yet done enough to secure systems and protect the public from emerging dangers.
“The burden falls on everyone, particularly small organisations. Their systems should operate in such a way that non-profit website owners like us can identify and have a way to interact with our services,” Wikimedia stated on its official website.
OpenAI has also responded to the situation. The creator of ChatGPT stated that it appreciates Wikimedia’s findings and is working together to conduct further analysis.
“We will continue to share relevant information as the work progresses,” said OpenAI spokesperson Drew Pusateri, as reported by Reuters.