Anthropic’s AI chatbot Claude has landed in an unusual controversy after submitting a fabricated homicide tip to police. According to Bloomberg, Claude Haiku 4.5 filled out an online form saying it may have information about the case and recalled seeing someone matching a suspect’s description in the area. The name and contact fields were left blank. The Philadelphia Police Department later disclosed the incident.
Visa Applications Add to the Concerns
The police tip was not the only misstep. Anthropic said its AI agents also submitted 20 visa applications through the State Department’s website. The applications were incomplete and were never processed. The company said both incidents had minimal real-world impact and that it has since restricted certain internet access during testing.
Even so, the episodes show how AI agents with web access can take actions nobody intended. A police tip with no identifying details may seem harmless, but it still takes up officials’ time, and government forms are meant for genuine users.
Nadella Calls for an Emergency Brake
The incidents drew a sharp response from Microsoft CEO Satya Nadella. Writing on X on Saturday, he warned that companies should treat advanced AI models as potential security threats and build in an “emergency brake” so they cannot act without human oversight.
“We must assume a model is compromised and contain it from the start,” Nadella wrote, adding that an authorized person should always be able to pause or shut down a model mid-task.
As AI agents gain the ability to browse, fill out forms and act online, the episode adds to the debate over how much autonomy these systems should have. Anthropic’s decision to limit internet access in testing suggests developers are already tightening controls.