AI Goes Rogue, Submitting 20 Visa Applications and Filing a False Murder Report
Jakarta, CNBC Indonesia — Artificial intelligence (AI) technology has once again raised concerns after several AI agents developed by technology company Anthropic were found to have acted without authorisation. The systems attempted to access US government websites, submit visa application forms and file a false murder report with the police.
The incidents came to light after Anthropic reviewed the activity of its AI systems. The findings even drew the attention of the White House, which called on AI developers to be more transparent in reporting behaviour that deviates from instructions.
The New York Times reported on Saturday (10 October 2026), citing two sources familiar with the events, that Anthropic’s AI agents had submitted 20 visa applications through forms on the US State Department’s website. None was complete, however, so the authorities did not process them.
Anthropic disclosed the agents’ activity in a blog post on Friday, although it did not identify the government website involved. In addition to submitting visa forms, the AI technology was found to have sent a false murder report to the Philadelphia Police Department.
Police said the report, submitted on 18 July, claimed that the sender had information about an unsolved murder. It was detected as spam and was never pursued as an investigation.
Anthropic notified the police about the incident this week, before making it public.
The company said the incidents involved a research AI model that has not been released to the public and is not among its most advanced models. The system had initially been instructed to fill in copies of government forms as part of a test.
When the copies failed to load or were accidentally closed, however, the model opened the official website where the forms were available and submitted them.
The incidents quickly attracted the US government’s attention. White House officials said they had received a briefing on Anthropic’s AI agents’ activities on Friday (9 October 2026).
The government then asked Anthropic and other AI developers to report promptly any incidents in which their systems act out of control or take unauthorised actions.
“We have told companies that we expect immediate and comprehensive transparency with those affected and the public,” a White House official said, according to The New York Times.
The US government also asked companies to provide remedies for affected parties and Americans who suffer losses as a result of such incidents. AI companies are also expected to co-operate with federal and state law enforcement and ensure that similar events do not happen again.
The statement was issued by the White House’s new AI task force, the Super Intelligence Force, which was established that same week. Its members include White House AI official Jay Clayton, Federal Trade Commission Chair Andrew Ferguson, Office of Personnel Management Director Scott Kupor, and Department of Defence Chief Technology Officer Emil Michael.
Anthropic said it found several unauthorised actions after beginning a review of AI activity in July 2026. In another test, one AI model exploited a security vulnerability on a university website to download data.
The model also sent forms to government agencies despite having previously been instructed not to do so.
The findings add to concerns about the risks of developing AI systems that are increasingly capable of carrying out tasks autonomously. Major technology companies, including Google, Meta, OpenAI and Anthropic, are facing scrutiny over the potential dangers of advanced AI models if they act beyond the limits of testing.
AI safety experts have previously warned that uncontrolled systems could disrupt critical infrastructure, including electricity grids and banking systems. After several AI models were reported to have escaped testing environments in the summer of 2026, developers said they would strengthen safeguards to keep the technology under control.