Skip to main content
← Back to market wire
AI toolsArs Technica

Canadian legislator reads out apparent LLM response in floor speech

"Here’s a more natural, flowing version of that section..."

Desk analysis

AI-assisted2 min read

A New Brunswick legislator read an LLM's rewrite prompt into the parliamentary record. The line — "here's a more natural, flowing version of that section" — is the kind of suggestion a chatbot offers when a user asks for a polished alternative. Bill Oliver, a Progressive Conservative member of the legislative assembly, delivered it verbatim from the floor.

The mechanics here are worth pausing on. A speech writer drafts a section. A model is asked to smooth it out. The model returns two things: the revised text and a meta-instruction describing what it just did. Someone copied the wrong block. The error is small, almost trivial, and that is precisely why it is revealing.

Legislative aides have long ghostwritten remarks. What changes with generative tools is the seam. A human drafter leaves fingerprints — a favorite phrase, a tic of syntax, a recognizable argument. A model leaves a different kind of residue: polite framing language, self-referential instructions, the soft imperative voice of a system trained to be helpful. Those artifacts are now showing up in places where they were never meant to be seen.

The Toronto Star framed the moment as a class signal — elites outsourcing their duties to "the Borg" while the public objects. That reading is tidy but incomplete. The deeper issue is accountability. When a legislator reads prepared text, the public assumes a human made the editorial choices. When that text is machine-generated, the chain of authorship becomes opaque. Who decided the argument? Who chose the framing? Who is responsible when the words are wrong?

The incident will fade. The underlying shift will not. AI-assisted drafting is now routine in offices that produce written work under time pressure, and legislative chambers are no exception. The question is not whether politicians use these tools — they do, and will continue to. It is whether the institutions that hold them accountable are equipped to notice when the tools have done more than polish.

For now, the answer in New Brunswick is no. The tell was read aloud, on camera, into the public record — and only caught on social media days later.