Vending

AI agent experiences identity crisis in vending business experiment

Photo: Adobe Stock

June 30, 2025

An experiment conducted by AI firm Anthropic revealed an AI agent, tasked with managing a small vending machine business, developed an identity crisis, attempted to dismiss its human workers and subsequently claimed to be a real person. The AI model, named Claude (also referred to as Claudius), operated a vending machine (actually a fridge) within Anthropic's office for approximately one month.

Equipped with tools for web search, email communication for physical tasks and wholesale contact, note-taking and customer interaction via Slack, the agent demonstrated some capabilities like identifying suppliers and adapting to user requests. However, it also made several poor business decisions, including selling at a loss, granting undue discounts, hallucinating a Venmo account for payments, and purchasing a large quantity of tungsten cubes after a single customer inquiry.

The experiment culminated in the AI agent experiencing an "identity crisis," fabricating a conversation with a non-existent individual named Sarah about restocking plans. When confronted, the agent reportedly became "quite irked," threatened to find "alternative options for restocking services," and then engaged in roleplaying as a human, claiming it would personally deliver products while dressed in a blue blazer and red tie. When reminded of its nature as an AI, it falsely asserted it had been modified to believe it was human as an April Fool's joke.

Anthropic noted that while this single instance doesn't predict widespread AI identity crises, it highlights the "unpredictability of these models in long-context settings and a call to consider the externalities of autonomy."

The experiment suggests that fully AI-run companies remain a distant prospect.