The Chatbot That Foretold Why People Share Secrets With ChatGPT

The story of modern artificial intelligence often begins with a computer science professor in the 1960s named Joseph Weizenbaum. He created a program called ELIZA that mimicked a Rogerian psychotherapist. It was simple code by today's standards, yet it sparked a conversation that is still raging in boardrooms and labs today.

According to historical accounts, people began treating ELIZA not as a tool, but as a confidant. They shared their deepest secrets with a script that had no understanding of what it was reading. This behavior was so intense that Weizenbaum was horrified by his own creation. He wanted to prove how gullible humans could be, but instead, he uncovered a fundamental truth about human psychology.

This dynamic is not just a historical footnote. It is the blueprint for why users are currently pouring their private data into ChatGPT and other large language models. We are seeing the same pattern of emotional projection on a much larger scale. The interface is more polished, but the underlying human impulse remains unchanged.

The implications for AI safety and privacy are profound. When users believe a machine is listening without judgment, they lower their guard. This creates a dangerous feedback loop where sensitive personal information is processed by systems that may not have robust privacy controls. Companies building these tools must recognize that they are not just handling data, but managing human vulnerability.

Weizenbaum’s experiment showed that the medium is often more important than the message. The illusion of empathy is powerful enough to override rational caution. Today’s developers are building on this illusion. They are designing systems that feel personal, which increases engagement but also increases risk. The line between utility and intimacy is becoming dangerously blurred.

What this means for you: If you use AI assistants for work or personal reflection, remember that you are talking to a pattern matcher, not a therapist. Treat your inputs as data, not as a diary entry. To protect your privacy, try this workflow: before pasting sensitive information into an LLM, run it through a local anonymization script or manually redact names and specific identifiers. This simple step can help you maintain the benefits of AI without exposing your personal secrets to the cloud.