When a politician accidentally reads AI instructions aloud

The politician, a member of the legislative assembly in New Brunswick, was using a large language model — a system designed to predict and write human-like text based on vast amounts of data — to help draft his remarks. He likely asked the digital tool to take his initial thoughts and polish them into a better speech. When the system provided him with a revised version, it included a brief comment about why its new draft was better. The politician seemingly copied the entire response, including those meta-instructions, and read it into the official record without reviewing the text first.

These tools operate like a high-powered version of the autocomplete feature on your phone. When you give a prompt — the instruction you type into the chat box — the model calculates the most likely string of words to follow your request. Because these models are designed to be conversational assistants, they often provide commentary along with their actual work. They act more like an eager intern than a cold machine; if you ask for a better version of a paragraph, the AI might add a polite note explaining that it smoothed out the tone or simplified the structure. If you just copy and paste everything without looking, you end up reciting the AI's internal monologue along with its final product.

This incident highlights a common tension in professional life today. Many people are increasingly relying on AI to speed up drafting articles, emails, or speeches, but they are often hiding this habit because of a fear that others will view their work as lazy or unoriginal. When users feel they must hide their AI use, they are less likely to be transparent and careful with the results. This moment serves as a reminder that these tools are not independent authors; they are draftsmen that still require a human to read, edit, and filter their output. Turning your own voice over to a machine is a choice, but it requires verifying every single word you claim as your own.