OTTAWA – Roughly one in five statements delivered on the floor of the House of Commons last year were flagged as written with the help of artificial intelligence, according to a Hansard-based analysis reported by The Economist, and the finding has split MPs over whether Parliament has a problem worth fixing.
The analysis ran comments and speeches longer than 100 words through Pangram, an AI-detection tool built by Brooklyn-based Pangram Labs, and found Canada’s Commons ranked second highest among five legislatures studied, a group that also included the United Kingdom, Australia, New Zealand and the United States, according to the Canadian Press.
Reaction split along lines that had little to do with party. Conservative MP Sandra Cobena told the Canadian Press “I think everybody’s using artificial intelligence,” calling it “a great research tool” that “has to be used with care.” Fellow Conservative Luc Berthold was blunter about how he uses it: “If the sentence is really good, I will probably put it in and use it.” Liberal MP Ben Carr drew a line between using AI “to collect your thoughts” and having it generate an entire speech, which he considers a problem. Liberal MP Greg Fergus was the most skeptical, describing himself as “an analog man” who does not think a machine belongs in a chamber where members are “talking about our laws.” The Conservatives and Liberals have no rule against AI-drafted statements, the NDP is drafting a formal policy and the Green Party leaves the choice to individual members.
What none of that coverage examined closely is how much weight the 20 per cent figure can actually bear, and that is where the real story sits for Canadians trying to judge whether their representatives are outsourcing their voices to a chatbot. Pangram’s own claims are strong: an independent 2025 University of Chicago study credited the tool with a near-zero false-positive rate and near-zero false-negative rate on longer passages. But the same research documented a specific weakness that matters enormously in this case. When text was adversarially edited or run through AI “humanizer” software to disguise its origin, Pangram misread it as human-written 86 per cent of the time, and the tool has historically struggled to tell a speech that is wholly AI-generated apart from one a member wrote personally and simply had AI polish or fact-check, flagging both the same way. A speech an MP drafted and asked a chatbot to tighten is not the same as a speech a chatbot wrote from scratch, and Pangram’s own documented limits mean the 20 per cent figure likely blends both without distinguishing them, a nuance missing from the wire coverage of the story.
That ambiguity exists in a governance vacuum. The House of Commons administration launched an internal AI working group in April 2024, rolled out online generative-AI guidance for members and staff in May 2025, and has a broader five-pillar AI strategy headed for approval this autumn, according to the Inter-Parliamentary Union’s own tracking of the institution’s AI adoption. None of that guidance requires an MP to disclose when a statement was AI-assisted, or draws a line between light editing and wholesale drafting, which is exactly why Berthold and Fergus can describe such different practices without either breaking a rule.
The gap fits a wider pattern. A federal public consultation on AI transparency rules, run by Innovation, Science and Economic Development Canada, closed Sept. 23, 2026, with no comprehensive federal AI law in place to answer the questions it raised. Parliament’s own AI strategy is still months from final approval. For a generation of voters who increasingly use the same AI tools in their own school and work, the unresolved question is not whether MPs use AI. It is whether Canadians have any reliable way to know when the words their representatives speak on their behalf are actually theirs.
Via The Economist and the Canadian Press. Read the original reporting: bnnbloomberg.ca.
