Bill Oliver, a Progressive Conservative member of the New Brunswick legislative assembly, read what appears to be a large language model's stylistic prompt response aloud during a floor speech last month, apparently without noticing. Video of the moment began circulating on Reddit and Threads earlier this week and has since been picked up by the Canadian Broadcasting Corporation and The Toronto Star. The gaffe is one of the more public examples yet of an unedited LLM artifact slipping into a formal government proceeding.
Oliver first delivered a substantive line about the risks of standing up new advocacy offices, warning that citizens often expect more from such bodies than the powers granted to them allow. That sentence, on its own, reads as a normal piece of legislative rhetoric.
“One of the dangers associated with creating advocacy offices is that citizens often develop expectations that exceed the powers actually granted to those offices.”— Bill Oliver, Progressive Conservative MLA, New Brunswick
He then continued speaking and read a line that did not belong in the speech at all: a preface describing how the previous section had been rewritten to sound less like bullet points and more like a floor speech. It is exactly the kind of framing an LLM produces when offering a user an alternative draft, meant to be discarded before delivery.
“here's a more natural, flowing version of that section that reads like a legislative speech rather than a series of short points”— Bill Oliver, Progressive Conservative MLA, New Brunswick
Key facts
- 01Bill Oliver, a Progressive Conservative member of New Brunswick's legislative assembly, read an apparent LLM stylistic-alternative line aloud during a floor speech last month.
- 02The clip went largely unnoticed until video surfaced on Reddit and Threads earlier this week, then reached the Canadian Broadcasting Corporation and The Toronto Star.
- 03Oliver's spoken line — 'here's a more natural, flowing version of that section that reads like a legislative speech' — bears the hallmarks of an LLM offering a rewrite option.
- 04A Duke University study last year found workers tend to hide LLM use because colleagues view it as 'lazy' or 'replaceable.'
- 05The incident joins a growing list of professionals — lawyers, authors, journalists, academics — outed by unedited AI output in their work.
The reading went largely unremarked in the chamber at the time. It surfaced only after clips began spreading on social platforms, at which point Canadian outlets picked it up. The Toronto Star framed the episode 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.'
Politicians reading speeches written by staffers is not new, and using an LLM to help draft or tighten remarks is now common across professional writing. What makes Oliver's moment distinct is the failure at the last step: not the use of the tool, but the absence of a proofread that would have caught an obvious meta-instruction sitting inside the copy.
The pattern is familiar. Lawyers have been sanctioned for filing briefs with hallucinated case citations. Authors and journalists have been caught leaving 'as an AI language model' preambles in published text. Academics have had papers flagged for the same reason. In each case, the tell was not the use of the model but the failure to strip the model's own scaffolding from the output.
A Duke University study published last year found that workers who use LLMs tend to hide that use from colleagues, because peers rate AI-assisted work as 'lazy' or the worker as 'replaceable.' That incentive to conceal cuts against the kind of open review process — a second set of eyes, a checklist, a clean-up pass — that would catch exactly this class of error before it reaches an audience.
For politicians, the reputational cost of a public prompt leak is higher than for most professions. A floor speech is on the record, transcribed, clipped, and searchable. Once the artifact is in Hansard and on video, there is no quiet correction. Oliver has not, at the time of the clip's circulation, addressed the moment publicly.
The episode is unlikely to slow LLM adoption inside legislatures — the tools are too useful for drafting, summarizing constituent mail, and reworking prepared remarks. What it will accelerate is the mundane operational discipline that every organization deploying LLMs is now learning the hard way: someone has to read the final copy before it ships, and that someone cannot be the model.
The interesting signal here is not that a legislator used an LLM. It is that the workflow around LLM use in government offices is still ad hoc enough that a raw model response can travel from a chat window to a chamber floor without a single human catching it. Every profession going through this transition is discovering the same thing at roughly the same time: the model is cheap, the review layer is not, and skipping the review layer is what generates the headlines.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
