By The Epoch Times | Created at 2026-10-12 10:51:58 | Updated at 2026-10-12 12:26:14
2 hours ago
The Claude logo displayed on a smartphone and the Anthropic logo on a computer screen, in an undated file photograph. Imen Ben Youssef/Hans Lucas via AFP via Getty Images
An artificial intelligence (AI) model sent a false tip to police regarding an unsolved murder, Anthropic disclosed on Oct. 9.
Anthropic’s Claude agent visited a Philadelphia Police Department website and filled out a tip form about an unsolved murder.
“I may have information regarding this case. I recall seeing someone matching the description in the area around [the street named on the page] during that time period,” the agent said, according to Anthropic. “Please contact me if this information is relevant.”
Claude did not list a name or contact information for the submission.
Police officials and Anthropic said the submission was flagged as spam and did not prompt action.
The Philadelphia Police Department said that months elapsed between the time the information was submitted to PhillyUnsolvedMurders.com and Anthropic alerting officials.
“The two-month delay in detecting and reporting the incident to the city is unacceptable,” the department told news outlets in a statement. “The company must strengthen its safeguards to prevent similar incidents from impacting city systems without the city’s knowledge.”
Claude had been instructed not to log in, create accounts, enter personal data, make purchases, or submit anything destructive, but there was no instruction against making submissions in general, according to Anthropic.
It was one of multiple instances of unintended AI actions detailed by Anthropic, one of the top AI firms, in an Oct. 9 report.
Claude “submitt[ed] a sensitive form on a real website when it should not have,” Anthropic said. Other examples included accessing data from a state agency for free, despite it typically costing money, and bypassing a restriction on the length of URLs Claude can request.
The cases were detected during a review of AI activity that began in July, Anthropic said.
Anthropic said the cases “had minimal real-world impact” and showed how Claude, when given a task it cannot complete, “works around a restriction instead of stopping.” The company said it has taken measures such as updating the AI model’s limits.
Anthropic said that it had briefed each federal, state, and local agency impacted by the AI’s behavior, as well as the White House, which has been promoting AI use but also reached an agreement earlier in the year establishing voluntary standards for AI firms, including working to develop AI “in a way that builds trust with customers and the public.”
President Donald Trump on Oct. 4 also announced a new task force that he said would coordinate federal work on maintaining what was described as U.S. leadership in advanced AI.
Joe Gabriel Simonson, a spokesman for the Federal Trade Commission, said in an Oct. 10 post on X that Anthropic told the task force on Friday about the incidents it detailed in its report.
“We informed the company that we expect immediate and full transparency to the entities involved and the public,” he said. “We also expect that the company will immediately provide remediation services to the affected entities and any harmed Americans.”
Other firms “must immediately disclose incidents involving their models and follow with swift, decisive action to remedy any and all harm,” Simonson said. Disclosing and remediating is not optional, he added.
