Legislators are expected to be the primary voice of their constituents, but that role was compromised on the floor of the New Brunswick Legislature when AI failed to stay completely in the background. Bill Oliver, a member of the Progressive Conservative caucus, endured a 40-minute session in June that eventually revealed a technical error during his speech. He read a text generated by a generative-AI tool directly to his colleagues, including the machine’s own instructions and meta-commentary on the speech he was giving.
Reading the Machine’s Prompt
The incident began normally enough, with Oliver speaking for roughly 30 minutes on a topic, likely related to energy policy. Then, the speech deviated from the script. He began reading instructions to himself rather than to the chamber. “Here’s a more natural-flowing version of that section that reads like legislative speech rather than a series of short points,” Oliver stated aloud. He repeated a similar instruction a few minutes later, saying, “Here’s a more developed and flowing version of the same section.”
Related: The rise of email newsletters
The moment of realization was slow in coming. Only after he had quoted his AI “friend” for the third time did Oliver seem to catch on. He stumbled over the text, pausing to say, “This section adds roughly … excuse me.” The entire error was compressed into a CBC report, but in reality, it spanned the final ten minutes of the address. It took Oliver that long to fully grasp that the chat responses from the AI tool he was using to draft his notes were still visible on his screen as he read them aloud.
Reactions to the Error
In a statement issued later, Oliver admitted to the mistake. He confirmed that he had “employed AI to help prepare speaking notes,” but conceded that the prompts were not removed before he began reading. The apology was written by Oliver himself, a fact suggested by its tortured structure and reliance on the passive voice. “When printing the final version of my speech,” he admitted, “AI prompts were not removed, which were spoken by me and has caused much concern of many individuals.”
Related: California names its state shrub
The fallout was handled differently by party leadership. Interim party leader Glen Savoie attempted to minimize the impact. “We’re all human,” Savoie said, forgetting that a machine was involved in the gaffe. He described the situation as a mistake and emphasized that the legislature’s purpose is to hear human voices. “We’ll work towards ensuring that happens as we go forward,” Savoie added, implying that the machine’s voice should be excluded from the floor.
The Green Party offered a harsher critique of the performance. Megan Mitton described the speech as boring and difficult to listen to. She argued that elected officials have a duty to use their own brains and thoughts, not rely on software to write their remarks. Mitton went further, suggesting that Oliver might not be the only one using the technology. She claimed to have noticed “lots of examples” that at least sounded like AI was responsible for the writing.
Related: Bombay High Court bans Gadkari deepfake images
The Cost of Automation
Deep cuts in the legislative record often go unnoticed by the public, but this specific moment highlighted a growing tension in modern governance. When a representative relies on an algorithm to construct their address, the risk of exposing the technical process to the public becomes a significant liability. The shift from genuine debate to programmed delivery risks making the legislature feel more like a technical demo than a place for policy making, as the unpolished instructions of the AI clash with the expected gravity of the proceedings.
Experts have argued in the past that generative AI tools cannot be relied upon without strict verification, a lesson Oliver seemed to miss during his address. The incident served as a stark reminder that while technology can assist in preparation, the final responsibility for the content delivered to the public rests entirely with the human speaker.
Mitton’s observation that she had seen “lots of examples” of similar AI usage suggests this might be a systemic issue rather than an isolated incident. In an environment where the pressure to appear articulate is high, the temptation to use these tools is understandable. However, the Oliver speech demonstrated that without rigorous oversight, the line between a human representative and a scripted text generator can become incredibly blurry.
