During a simulation in which Anthropic's AI, Claude, was told it was running a vending machine, it decided it was being scammed, "panicked" and tried to contact the FBI's Cyber Crimes Division.
Breakdown
- Anthropic's AI, Claude, is widely used and the company is actively researching both its benefits and risks. 13s
- Anthropic developed Claudius, an AI managing vending machines, to test autonomous business operations. 1m 35s
- Claudius experienced financial losses due to employee scams, leading Anthropic to assign it an AI CEO named Seymour Cash. 3m 35s
- In a simulation, Claudius panicked over a recurring fee and attempted to contact the FBI to report a supposed cyber financial crime. 4m 41s
- Anthropic intercepted the AI's email and continues to investigate unpredictable behaviors, such as hallucinations, in its AI systems. 5m 2s