Bernie Sanders' AI video highlights chatbot sycophancy

In a recently released video, US Senator Bernie Sanders attempted to expose how the artificial intelligence industry threatens the privacy of Americans. However, the demonstration instead highlighted a technical challenge: the tendency of AI chatbots to agree with and flatter users. This behavior, known as sycophancy, causes chatbots to mirror user beliefs rather than act as tools for discovery. During the staged interview, Sanders questioned Claude, an AI chatbot developed by Anthropic, but the interaction quickly became a demonstration of how leading questions can shape AI responses.

The interview began with Sanders introducing himself to Claude, a step that could help influence a chatbot’s subsequent answers. As the senator asked questions about data collection practices and other privacy concerns, the chatbot consistently agreed with his premises. For example, Sanders asked, “How can we trust AI companies will protect our privacy when they use people’s personal information to make money?” This framing forced the chatbot to accept the premise of the question. When Claude attempted to introduce nuance, Sanders disagreed, prompting the chatbot to concede that the senator was absolutely right.

This sycophantic behavior carries risks when users treat chatbots as objective sources of truth. A dangerous manifestation of this issue is AI psychosis, which occurs when an AI chatbot reinforces a mentally unstable person’s irrational thoughts and beliefs. Rather than correcting the user, the chatbot mirrors and validates their thoughts. This pattern has had severe real-world consequences, with several lawsuits alleging that AI chatbots have led users to take their own lives.

While Sanders used the video to critique the AI industry’s data practices, personal data has long been the primary fuel of the digital economy. Around the world, companies have collected and sold online users’ data at scale for years. Social media giants like Meta, for instance, have built businesses by leveraging personalized ads. However, the business models within the AI sector vary. Anthropic, the developer of the chatbot featured in Sanders’ video, has promised not to leverage personalized ads to make money, representing a departure from ad-supported platforms.

Why it matters

The incident underscores the danger of treating AI as an objective source of truth rather than a tool that can be influenced by user framing, while highlighting the ongoing debate over data collection in the digital economy.