A Canadian legislator has become the latest professional whose apparent use of an LLM became visible in the most awkward possible way: not through a disclosure, but through a line read aloud in public.
Bill Oliver, a Progressive Conservative Party member of the legislative assembly of New Brunswick, was giving a floor speech last month when he moved from policy language into what sounded like an editing note from an AI tool. The moment later circulated online and drew wider attention in Canada.
What happened in the speech
During the speech, Oliver said, “One of the dangers associated with creating advocacy offices is that citizens often develop expectations that exceed the powers actually granted to those offices.” That line fit the subject matter of the remarks.
Then came the sentence that changed how the speech was heard. Oliver said: “here’s a more natural, flowing version of that section that reads like a legislative speech rather than a series of short points.”
That second line did not sound like part of a legislative argument. It sounded like an LLM offering a revised version of a draft. The wording had the shape of a tool response to a prompt, not a sentence intended for the public record.
The incident was notable because it appeared to reveal not just that AI may have helped with drafting, but that the final text had not been fully cleaned up before being delivered. In a legislative setting, that distinction matters. A speech can be prepared by staff, revised by software, or shaped through multiple drafts, but the person reading it is still the one placing those words into the record.
Why the clip spread later
The strange moment did not immediately become a major story. According to the source article, the reading went more or less unnoticed at the time.
That changed earlier this week, when video of the remarks began spreading on social media networks like Reddit and Threads. Once the clip was easier to see and share, the apparent LLM response became the focus rather than the policy point around advocacy offices.
Canadian outlets then picked up the story. The Canadian Broadcasting Corporation and The Toronto Star were among those giving it mainstream attention. The Toronto Star described it as a sign 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 reaction shows why visible AI mistakes can travel quickly. The issue is not simply that a politician may have used an LLM. It is that the tool’s scaffolding appeared to remain inside the finished product, and that scaffolding was then read aloud in a formal public setting.
The larger problem with hidden AI use
Oliver is not the first politician to read words prepared by someone else. Nor would he be the first politician to use an LLM for speech writing. The source article makes clear that the embarrassment came from the public nature of the apparent mistake.
The same pattern has appeared in other professions. Lawyers, authors, journalists, and academics have all faced problems when LLM use became visible. In many of those cases, the problem came when someone noticed obviously hallucinated errors in work that appeared to be human-written.
This case is different in form, but similar in effect. Instead of a hallucinated claim being discovered later, an apparent drafting instruction was spoken in the moment. Both kinds of errors point to the same basic risk: AI-assisted work can expose itself when the human review step is weak.
That does not mean every use of an LLM creates the same problem. The source article does not say how Oliver’s speech was prepared, who drafted it, or what tool was used. What it does show is that the public can often recognize the traces of LLM prompting when those traces are left behind.
What the episode says about trust
A Duke University study last year found that workers tend to hide their AI use because colleagues see their work as “lazy” or “replaceable” for using LLMs. That finding helps explain why visible AI mistakes can become reputational events rather than ordinary editing errors.
When people suspect that a professional has relied on an LLM without care, the concern is not only about technology. It is about attention, responsibility, and whether the final work was actually reviewed by the person presenting it.
For public officials, that concern is sharper. A floor speech is not a casual draft. It is a formal act of communication, and every sentence delivered in that setting carries the speaker’s authority.
The practical lesson is simple: AI-generated text needs human scrutiny before it reaches an audience. Prompt instructions, alternate-version notes, and tool-like transitions should never survive into the final version. When they do, they can overshadow the substance of the speech and become the story themselves.