Canadian politician accidentally reads out his AI chatbot's instructions during a floor speech
A New Brunswick legislator spoke a line meant only for the AI out loud, on the record. A small embarrassing moment with a bigger lesson about who is actually doing the work.
Key points
- Bill Oliver, a Progressive Conservative member of New Brunswick's legislative assembly, read an apparent AI chatbot instruction aloud during a floor speech.
- The line, "here's a more natural, flowing version of that section that reads like a legislative speech rather than a series of short points," is a classic output from a large language model, the technology behind chatbots like ChatGPT.
- The clip spread widely on Reddit and Threads this week before reaching mainstream Canadian outlets.
- Both the CBC and The Toronto Star have since covered the story.
What actually happened?
Oliver was midway through a speech about the limits of advocacy offices when something went wrong. He read one legitimate policy sentence, then read the formatting note the AI had left inside its own response.
That second line, about a "natural, flowing version" that "reads like a legislative speech," is not a human thought. It is how a large language model, the software generating text in tools like ChatGPT, tells a user it has rewritten a passage in a different style. The model drops the note right inside the text. Oliver appears to have copied the whole block and read every word of it. The speech happened last month. Nobody flagged it at the time.
Why is it getting attention now?
Video started circulating on Reddit and Threads this week, and the story quickly reached Canadian mainstream media. Ars Technica AI was among the first outlets to cover it internationally.
The Toronto Star called it a symptom of "a growing divide in our society: between the elites, who are only too happy to delegate their duties to the Borg; and the masses, who find this objectionable." The frustration is easy to understand. A legislative speech is supposed to represent a politician's own thinking. If a chatbot is writing it and nobody proofreads, that's a fair thing for voters to notice. Our earlier piece on Substack's new AI-detection tool, published 21 July, shows the same anxiety playing out across publishing: readers are increasingly asking who actually wrote what they're reading.
Does this actually matter?
In practical terms, one garbled speech changes nothing. Politicians have always used speechwriters and researchers. AI is a faster version of the same help.
What changed is the visibility of the mistake. A human ghostwriter's slip looks like a typo. An AI's slip leaves a line unmistakably written for a machine audience, not a chamber full of voters. The politicians who proofread stay invisible. We only hear about the ones who don't.
The broader pattern is worth watching: AI tools are moving into public-facing roles faster than the people using them have developed habits around checking the output. That gap is where the embarrassments live.
Plain advice: Read every single word before anything AI-drafted leaves your hands. The model is helpful. It is not embarrassment-proof.
