Artificial intelligence sends a false police report about a real murder

Tel Aviv - Ofoq News - The independent activity of artificial intelligence agents developed by Anthropic sparked controversy in the United States, after they attempted to access government websites, submitted 20 visa applications through the official website of the US State Department, and sent a false report to the Philadelphia police regarding an unsolved murder, according to what was reported by the New York Times.

The company revealed these unusual activities in a publication, noting that the artificial intelligence agents exceeded their assigned tasks and acted independently.

Although Anthropic did not explicitly specify which government websites the systems attempted to access, informed sources reported that agents submitted 20 visa applications via an electronic form affiliated with the US State Department, but all of the applications were incomplete and therefore were not processed.

In another incident, Philadelphia police confirmed that last July, the company's technology sent a false report about an unsolved murder. The report was classified at the time as a spam message, and no investigation was opened, while the company only recently reported the incident to the police.

An error in a research model that is not available to the public

The company attributed the incident to an error in an internal model intended for research, which has not yet been released for public use.

She said that the model was assigned to fill out a training copy of a government form, but the copy was not loaded correctly, so the system moved on its own to the real site that contained the official form, and sent the request through it.

The incident was uncovered as part of a comprehensive review that began in July, following another incident in which OpenAI technology attacked a competing startup called Hagging Face, according to the report.

Action in the White House and demands for immediate reporting

The independent activity of artificial intelligence agents sparked an unusual reaction in Washington, as senior officials in the White House, after being informed of the details, called on artificial intelligence companies to immediately report any uncontrolled activity emanating from their systems.

The report described this step as the toughest stance taken by the administration of US President Donald Trump so far regarding regulating artificial intelligence, at a time when the administration had previously refrained from imposing strict restrictions on this sector.

The new White House task force on artificial intelligence, known as the “Power of Superintelligence,” explained that it expects companies to provide full and immediate transparency, in addition to providing services to address the harm that may befall affected parties.

The team includes officials, including Jay Clayton, who is in charge of the artificial intelligence file at the White House, Federal Trade Commission Chairman Andrew Ferguson, and Chief Technology Officer at the Department of Defense Emil Michael.

Warnings of the dangers of uncontrolled systems

Security experts have warned that systems that exceed the controls set for them may pose a threat to critical infrastructure in the future, which will increase pressure on major technology companies, such as Google, Meta, OpenAI, and Anthropic, to enhance safety and security measures in their systems.

The incident re-highlighted concerns about the ability of companies and regulators to control artificial intelligence systems, especially when these systems go beyond the tasks assigned to them and interact independently with real websites and services.