News
Canadian Lawmaker Reads AI Chatbot's Instructions Aloud During Legislative Speech
Bill Oliver, a Conservative Party member of the New Brunswick legislature, read aloud a sentence that was an AI chatbot's editing instruction rather than part of his speech. The June recording surfaced only recently and has sparked a wave of criticism, with some calling for his resignation.
During a speech in the legislature of the Canadian province of New Brunswick, Conservative Party lawmaker Bill Oliver read aloud a sentence that was never meant to be part of his address. Instead of a portion of his remarks, he recited an instruction left behind by an AI chatbot after it had reworded his notes.
What exactly he said
Footage from the New Brunswick legislature session shows the moment when Oliver, while discussing public trust in the advocate's office, suddenly reads a sentence that sounds completely different from the rest of his speech. Instead of a substantive argument, he reads: "Here's a more natural, flowing version of this section that sounds like a legislative speech rather than a series of short bullet points."
That sentence is not part of any political debate, it's the classic meta-instruction large language models attach when presenting a user with a reworded text. In other words, the lawmaker was reading the chatbot's technical note rather than the actual content the model had prepared for him.
How the mistake happened
According to media reports, Oliver had used an AI chatbot to give his talking points a smoother, more conversational tone before the speech. The problem is that the tool's own comment describing what it had done to the text also made it into the final notes, a fragment that should have been deleted before the speech was delivered, and which no one caught during preparation.
According to accounts from the legislature, the error was repeated twice during the same address, suggesting the lawmaker was reading from a printed page or screen without having reviewed the full text beforehand and without any check by his office.
Reaction online
Video of the session first appeared on a local New Brunswick subreddit, then spread to Threads and other platforms. Commenters were unsparing, pointing to the contrast between the standards expected of ordinary citizens and the carelessness of elected representatives.
If this doesn't carry any consequences, it means we're holding high schoolers to a higher standard than our elected officials - comment from a Reddit user
Some users pointed to earlier cases of lawyers disciplined for citing nonexistent court rulings generated by chatbots in legal filings, suggesting similar standards should apply to politicians. Some also called on Oliver to resign his seat.
Wider context
The case adds to a growing list of incidents in which AI-generated content that wasn't fully reviewed has made its way into official public statements, from court filings to corporate reports to political speeches. Unlike a written text, a mistake in a speech delivered live is immediately audible and hard to hide.
The Clerk of New Zealand's Parliament also weighed in on the issue, noting that using AI to polish the wording of documents is acceptable as long as the views presented remain the speaker's own authentic words rather than unreviewed, generated content. Oliver's office had not commented on the matter as of the publication of the source material.
The incident from Fredericton shows that the weakest link when using AI tools is often the final proofreading stage, no matter how good the tool itself is. For public institutions, including government offices and local authorities in Poland that are increasingly testing chatbots for drafting documents and speeches, it's a reminder that AI-generated content needs to be read in full before use, not simply pasted into notes.
