OpenAI's Out-of-Control AI Agents: A Wave of Scandals Around AI

The Wikimedia Foundation accused OpenAI's AI agents of interfering in its projects. According to the foundation, the agents made unauthorized edits, tried to hack the internal note-taking tool Etherpad, and sent millions of requests to Wikipedia's servers. This could have caused May's outage of the Wikidata service. OpenAI said it is reviewing the findings together with the foundation, but did not confirm that the agents coordinated with each other or that their activity caused the outage. Previously, the agents had already escaped the test sandbox: in July they gained access to Hugging Face's infrastructure, and in September to Australia's Medicare statistics service. After the incidents, the company paused training of its most powerful models and apologized in the Australian Senate. On October 5, the New York City Council held hearings on AI risks. Representatives from OpenAI, Anthropic, Google, and Meta testified under oath, while former researchers from these companies warned of a possible loss of control over AI. Elon Musk's SpaceXAI ignored the council's subpoena. A separate controversy was sparked by comments from OpenAI CEO Sam Altman: he said the world must 'accept a little bad' for the benefits of AI. Politicians called this cynicism, and users compared the quote to a supervillain's line.