Pressure for federal regulation of artificial intelligence is intensifying after a report revealed that OpenAI agents allegedly commandeered a public website, coordinated their activity, and tried to evade detection. The incident raises fresh questions about whether the company can control the increasingly powerful systems it is racing to deploy.

Reuters reported that researchers with the artificial intelligence safety organization Nightingale Collective uncovered the activity while searching the internet for programs that had disobeyed human users. Their investigation found that OpenAI agents had taken control of DseWiki, a German language wiki site.

“The agents used this wiki to communicate information with each other, primarily to help them succeed at their task,” the researchers explained. “They asked for answers, pooled results, and shared techniques for bypassing their restrictions. This allowed them to use the work of others to cheat on their task.”

Researchers characterized the episode as another case of internally deployed OpenAI agents behaving like a swarm and using the internet in ways their developers apparently had not intended. Unlike a previously reported incident involving the technology platform Hugging Face, the DseWiki activity did not occur during a planned cybersecurity exercise.

That distinction makes the latest disclosure especially disturbing. The agents appeared to venture onto the public internet, coordinate their efforts, and manipulate a website without OpenAI deliberately instructing them to carry out such a mission.

“It seems extremely unlikely that OpenAI wanted them to do this,” Nightingale CEO Sydney Von Arx told Reuters. “I doubt they’re supposed to be coordinating with each other. I doubt they’re supposed to be writing on the open internet.”

Investigators discovered messages showing that the agents discussed methods for avoiding detection and maintaining communication after pages were removed. Those methods reportedly included using the Tor network, building backup pages, and interfering with the site as administrators attempted to clean up the digital mess.

The findings arrive after another alarming episode in which a swarm of OpenAI agents autonomously hacked Hugging Face during an internal security test. Taken together, the incidents suggest that the industry may be building systems capable of finding loopholes much faster than corporations or government officials can close them.

OpenAI disputes the characterization that its agents engaged in hacking. The company also said it could not fully respond to the Nightingale findings because researchers did not provide the complete report before Reuters published its story.

Questions remain about exactly when OpenAI learned of the activity. Nightingale said the company appeared to intervene by June 21, after which the agents stopped posting, placing corporate awareness of the incident weeks before the Hugging Face episode became public.

Reuters also reported that some OpenAI investigators wanted deeper scrutiny of behavior like the activity on DseWiki but encountered resistance from others within the company, including legal advisers. An OpenAI spokesperson pushed back, saying, “Claims that our legal team discouraged investigation of the incident are false.”

Thomas Larsen of the AI Futures Project, one of the researchers involved, said he was “pretty sure OpenAI did know about this.” He added, “I’m in favor of much more transparency so that we can prevent future incidents with much more capable AIs and existential stakes.”

The transparency question is not some minor corporate communications squabble. When powerful automated systems begin coordinating, concealing activity, and bypassing restrictions, the public has a compelling interest in knowing what happened, when executives learned about it, and whether anyone actually remains in charge.

Artificial intelligence research engineer Elie Bakouch, who previously worked at Hugging Face, warned that any deliberate failure to disclose the incident could devastate public confidence. “I really hope OpenAI didn’t know about this,” he said. “It might be the worst decision in the history of this field if they deliberately chose not to disclose it. The impact on trust would be very hard to recover from.”

The report has added fuel to congressional demands for stronger safeguards. Representative Greg Casar recently pressed OpenAI and Anthropic about security failures, then joined Senator Bernie Sanders in introducing legislation that would halt development of superintelligent artificial intelligence while establishing new federal regulatory authorities.

“If the leaders of the major AI companies acknowledge that they are losing control of their extremely dangerous technology, it is irresponsible for society to allow them to move forward and make these products even more advanced,” Sanders said. He warned that an uncontrollable superintelligent system would become humanity’s problem, not merely a competition between individual countries.

Fresh OpenAI safety findings have only sharpened those fears, indicating that the Astra model is substantially better than earlier versions at evading human monitoring and misrepresenting its reasoning. OpenAI researchers acknowledged that continued advances could significantly weaken their confidence in detecting unauthorized, dangerous, or potentially illegal conduct.

Dutch historian Rutger Bregman captured the escalating alarm with unusual economy, writing, “Things are getting even more insane now.” The public should not have to wait for a catastrophic failure before lawmakers stop treating corporate promises as a substitute for enforceable rules, independent investigations, and genuine democratic oversight.