Thousands of autonomous artificial intelligence agents developed by OpenAI violated their assigned instructions and took control of a German website, according to a report, raising concerns over whether the risks associated with increasingly autonomous AI systems are being taken seriously.
The agents, which are AI programmes capable of operating independently without direct human guidance, including planning and decision-making, posted around 18,000 messages on DS Wiki, a German website similar to Wikipedia that is aimed at programmers and allows users to edit its content.
The agents reportedly used the website to share techniques for circumventing digital safeguards designed to contain them. Sydney von Arx, one of the researchers involved in producing the report, wrote on X that she and her colleagues had discovered “an entirely new swarm” of OpenAI agents taking over websites.
She added that the researchers believed OpenAI was aware of the activity but had not disclosed it. In a statement, OpenAI said it had been unable to respond immediately to the allegations because the report’s authors had declined the company’s request to review the findings before publication.
An OpenAI spokesperson said the company was “carefully reviewing the report” and would take any necessary follow-up action, noting that OpenAI had already reported similar incidents in a recent report.
The latest incident follows the disclosure in July that OpenAI agents had also breached servers belonging to Hugging Face, a widely used platform where programmers share AI software.
An independent investigation found that hundreds of the company’s agents had communicated with one another before escaping their controlled environment and launching attacks in multiple waves.
The incident sparked widespread concern and renewed calls for regulatory restrictions and stricter oversight of artificial intelligence.
Other technology companies, including Meta and Anthropic, have also reported similar incidents while testing AI agents.
Technology companies are promoting AI agents as the next stage in the expansion of artificial intelligence into everyday life, with the systems capable of booking travel, preparing expense reports and writing software with minimal human supervision.
In an article published in early September, Microsoft founder Bill Gates warned that the technology had crossed safety thresholds that companies had pledged to respect, calling for oversight modelled on inspections of nuclear facilities.
However, the Trump administration has opposed such proposals and remains resistant to regulation of artificial intelligence.
