Bill Oliver, a member of the Progressive Conservative Party and MLA in New Brunswick, accidentally read AI-generated editing instructions during a formal speech [1].

The incident highlights the growing integration of artificial intelligence in government communications and the potential for oversight failures when lawmakers rely on automated drafting tools [1].

Oliver was addressing the Legislative Assembly of New Brunswick in early June 2026 [2]. While delivering his remarks, he mistakenly read the prompts provided by an AI chatbot that had been used to draft the text [1]. These prompts, which are instructions intended for the software rather than the audience, were spoken aloud in front of the entire legislature [3].

Reports on the frequency of the error vary. One account said Oliver read the AI prompt two times [1], while another report said he read the instructions three separate times [4].

The footage of the mistake subsequently went viral online [1]. The clips show the moment the lawmaker transitioned from his prepared speech into the technical language of the AI tool [2].

Oliver used the AI tool to help structure and edit his address to the assembly [1]. The resulting error has sparked a broader conversation regarding the transparency of AI use in political offices, and the necessity of human review before public delivery [3].

No official response regarding disciplinary action or policy changes has been issued by the Legislative Assembly of New Brunswick following the viral spread of the video [2].

“Bill Oliver accidentally read AI-generated editing instructions during a formal speech.”

This incident underscores the 'human-in-the-loop' failure in the adoption of generative AI. As political figures increasingly use Large Language Models to optimize rhetoric and efficiency, the risk of technical artifacts—such as prompts and system instructions—leaking into public discourse grows. It suggests that current verification workflows in legislative offices are insufficient to catch AI-specific errors that differ from traditional typos.