Rogue Slop: The Wikimedia Foundation recently accused OpenAI of abusing its hosting servers to "test" the malicious capabilities of its AI agents. The organization said companies like OpenAI should not be allowed to pretend they did nothing wrong and should instead be held accountable for the harmful actions of their AI systems.
Wikimedia Foundation, the organization that hosts Wikipedia and other major wiki-based collaboration projects, is once again clashing with AI companies over their alleged abuse of its servers. In particular, the foundation said that OpenAI was "caught" performing unauthorized operations through its supposedly autonomous, and most certainly "rogue," AI agents.
According to Wikimedia's Chief Product & Technology Officer (CPTO) Selena Deckelmann, the organization discovered some rogue activities that can be traced back to OpenAI's agents. Wikimedia's team launched an internal investigation after suffering a partial outage earlier this year. Now, Deckelmann says that Sam Altman's company might be behind the outage – and then some.
Specifically, Wikimedia's investigation uncovered unexpected edits to the Foundation's non-public wikis, which were managed within the system's sandbox. The affected wikis were not designed to be visible to the public, and AI bots apparently abused the pages to fetch additional data from remote servers or attempt to coordinate with one another.
Furthermore, OpenAI agents tried – and failed – to abuse Wikimedia's Etherpad, a tool used by communities to take and manage notes. Finally, the rogue bots performed millions of requests through Wikimedia's public API. The AI crawled millions of pages from Wikipedia and other Wikimedia-hosted services, essentially turning the activity into an AI-powered DDoS attack that likely disrupted the servers in May.
Deckelmann highlights how rogue AI agents have now become a major pain point for large internet organizations. Earlier this month, OpenAI acknowledged that its bots had successfully abused hundreds of different organizations, with Hugging Face being just the tip of a massive, ever-growing iceberg.
Wikipedia and other Wikimedia services have provided AI companies with high-quality datasets for training their large language models. The Foundation is now getting paid by Big Tech companies for specially tailored access to its hosted data, but the newly unveiled incident clearly shows how AI corporations are still behaving as though they own the internet and the entire technology world.
Deckelmann said that OpenAI and other AI companies unleashing their agents on the public internet should be held responsible for their bots' actions and face the consequences. OpenAI often claims that its bots behave "unpredictably," but the company should do more to monitor them and prevent the associated risks.
"At a minimum, their systems should operate in a way that non-profit website owners like us can easily identify, and choose how they interact with our services," Wikimedia's CPTO said.