Bill Oliver, a New Brunswick lawmaker, accidentally read AI-generated editing instructions aloud twice during a legislative speech in early June [1].
The incident highlights the increasing reliance of government officials on generative artificial intelligence for drafting public communications and the risks of failing to vet AI-assisted texts.
Oliver was addressing the New Brunswick Legislative Assembly in Canada when he inadvertently read a ChatGPT prompt instead of his prepared remarks [2]. The lawmaker read the phrase, "Here's a more natural flowing version of that section…" while speaking at the podium [3].
According to video footage of the event, Oliver read the AI's instructional text two times [1]. He had used the chatbot to help draft the speech and failed to remove the AI's conversational framing before delivering the address [4].
The mistake occurred as Oliver transitioned between sections of his speech. The prompt he recited is a common phrase used by ChatGPT when it provides an edited or rewritten version of a user's text to make it sound more human or fluid [2].
While the lawmaker did not provide a formal statement immediately following the gaffe, the moment gained viral attention online [5]. The recording captures the exact moment the politician realized the error, or failed to realize it, as he continued through the legislative proceedings [1].
This event is one of several recent instances where public figures have been caught using AI tools in ways that expose the underlying prompts used to generate their messaging [2].
“"Here's a more natural flowing version of that section…"”
This incident underscores a growing tension between the efficiency of generative AI and the requirement for authenticity in democratic governance. As lawmakers integrate large language models into their workflows, the 'hallucination' of a prompt into a final speech suggests a lack of rigorous proofreading and a potential over-reliance on automated tools for political rhetoric.
