Originality.ai finds AI traces in New Brunswick legislative transcripts

Originality.ai says a June 9 speech by New Brunswick politician Bill Oliver showed signs of AI drafting residue, then prompted a broader review of 204 Hansard transcripts. The study found 22.9% of eligible legislative speaking turns in the latest 12 months were classified as Likely AI, raising questions about disclosure and review in public communications.

Why it matters: - The findings suggest AI-generated drafting traces may be showing up in public legislative records. - Originality.ai says those traces raise questions about review, disclosure and accountability in official communications. - The study also offers a broader snapshot of how often AI-like text patterns appear in New Brunswick legislative speech.

What happened: - Originality.ai published a study on what it calls “AI residue,” meaning drafting instructions, placeholders and other traces of AI assistance left in finished content. - The study was prompted by a June 9, 2026 speech by New Brunswick politician Bill Oliver during debate on Bill 46. - In the official legislative video, Oliver appeared to read drafting language aloud, including: "Here's a more natural flowing version of that section." - Originality.ai transcribed the roughly 34-minute speech, split it into 24 timestamped blocks and classified the full transcript and 19 of the 24 blocks as Likely AI using Originality.ai with a 15% AI Allowance. - A Likely AI classification does not prove AI wrote the speech, and detector confidence scores do not equal a percentage of words written by AI.

The details: - Researchers reviewed 204 official New Brunswick Hansard transcripts covering March 2022 through May 13, 2026. - The dataset included 18,023 pages and more than five million words. - After limiting the analysis to English-language speaking turns longer than 100 words, the team evaluated 8,175 speaking turns. - Over the latest 12 months, 499 of 2,181 eligible speaking turns, or 22.9%, were classified as Likely AI. - Likely AI classification rates increased each year included in the analysis. - Among 28 politicians with at least 30 eligible speaking turns, 14 had classification rates of 25% or higher and five had rates of at least 50%. - Bill Oliver’s rate was close to the group average, with nine of his 34 eligible turns classified as Likely AI versus 26.1% across the 28 qualifying politicians. - Statements of Condolence and Congratulations had the highest classification rate at 59.3%. - Statements by Members were classified as Likely AI at 25.3%, while Statements by Ministers were at 21.5%. - Oliver’s June 9 speech was outside the broader analysis window because the official transcript was not yet available at the study cutoff. - Researchers collected 204 official New Brunswick Hansard transcripts and kept the original floor-language column. - Consecutive remarks by the same politician were combined when no other identified speaker intervened. - The final record produced 17,582 speaking turns from 76 politicians. - Each complete turn was scanned once using Originality.ai Lite with a 15% AI Allowance. - Scores of 50% or higher were classified as Likely AI. - Hansard is an edited official record rather than a verbatim transcript, so the findings should be read as detector classifications within that record, not proof that individual politicians used AI. - Supporting materials include the full study, official Legislative Assembly webcasts, the June 9 video excerpt, and Originality.ai's analysis of the full speech.

Between the lines: - The June 9 speech served as a case study, but the larger conclusion comes from a separate review of legislative records. - The pattern suggests AI-assisted drafting may be more visible in prepared remarks than in spontaneous debate. - Because Hansard is edited, the study reflects detector output on an official record, not a forensic determination of authorship.

What’s next: - Originality.ai’s findings point to more scrutiny of public-sector speech, transcription practices and disclosure norms. - The company’s research suggests future reviews may focus on whether AI drafting traces are becoming more common in formal political communications.

The bottom line: - A single on-air slip led Originality.ai to a wider warning sign: AI-like residue appears in a meaningful share of New Brunswick legislative speaking turns.