When New Brunswick legislative member Bill Oliver stood before his peers to address public confidence in the Office of the Advocate, he did not just deliver a speech. He delivered an accidental confession regarding modern political workflows.
Reading directly from his notes, Oliver stated, "Public confidence in the office of an advocate matters," before seamlessly transitioning into a mechanical meta-commentary. "Here's a more natural, flowing version of that section that reads like a legislative speech rather than a series of short points," he intoned, entirely oblivious to the fact that he was reading aloud the conversational filler of a generative large language model.
The room did not erupt. The legislative broadcast carried on. Yet the digital ecosystem quickly seized upon the clip, turning a routine assembly proceeding into a viral testament to administrative malpractice.
This slip is funny on the surface. Beneath the humor lies an uncomfortable truth about how the people writing our laws use technology.
The Illusion of Preparedness in Modern Governance
Elected officials operate under crushing administrative burdens. Between committee meetings, constituent emails, and endless policy briefings, time is a scarce commodity. Delegating the drafting of speeches to junior staffers, communications officers, or external consultants has been standard practice for centuries.
Delegating that drafting entirely to machine learning algorithms without a human pair of eyes checking the output is a relatively new hazard.
When a politician stands up to read text they have never actually looked at, the contract of representation is quietly broken. The public assumes that an elected official has internalized, or at least read, the words coming out of their mouth. When an automated dialogue preamble slips past the final line of defense and echoes through a parliamentary chamber, it reveals a hollowed-out process.
Nobody proofread the document. Nobody cared enough to check the formatting. The text was generated, pasted into a teleprompter or a paper binder, and spoken aloud as absolute truth.
The Mechanics of Lazy Delegation
Using language models for administrative tasks is widespread across corporate and government sectors. The workflow usually looks identical. An overworked staffer feeds a handful of bullet points into a chatbot with a command like, "Here are the facts on the advocate's office. Make it sound professional."
The system responds with conversational pleasantries. It acknowledges the instruction. It offers transitional phrasing designed to chat with the user, not to be broadcast to a legislature.
[User]: Summarize these points into a speech.
[Model]: Here is a more natural, flowing version of that section that reads like a legislative speech...
[Staffer]: (Copies and pastes the entire block without scrolling to check the top)
[Politician]: (Reads the paste verbatim into a live microphone)
This pipeline depends on total passivity. It requires the staffer to abandon basic editorial hygiene and the politician to surrender any remaining editorial skepticism. When both fail simultaneously, the machinery of state effectively runs on autopilot, guided by automated text predictors that possess zero understanding of civic duty, constitutional nuance, or local context.
Accountability in the Age of Automated Text
Judges in various jurisdictions have already begun penalizing attorneys who submit court filings riddled with hallucinated case law or unedited chatbot markers. Legal professionals face sanctions, fines, and professional suspensions for failing to verify documents bearing their signatures.
Legislators operate under a different set of professional expectations.
When an attorney submits unvetted automated text to a court, it is treated as a breach of professional competence. When a politician reads unvetted automated text into a parliamentary record, it is treated as an amusing internet meme. That disparity points to a double standard in how we regulate intellectual rigor across public offices.
If lawmakers expect citizens and professionals to adhere to strict verification standards, the production of legislation and public commentary demands an equivalent level of human oversight. Blindly trusting an algorithm to manufacture political rhetoric turns representation into a pantomime.
The viral moment in New Brunswick will fade from headlines. The underlying habit of outsourcing thought to software will not. Every time an official relies on predictive text to bypass the hard labor of critical thinking, democracy grows a little more automated, a little more detached, and a lot more hollow.
