No standard used to evaluate AI, in Canada or elsewhere, treats consciousness as a property being tested for — every one of them measures what a system does, not whether anything is felt.
Short answer
No Canadian or international standard treats consciousness as a property AI is evaluated for, and that is not an oversight. Every framework in use, including the one Canada’s own privacy regulators point to, measures functional properties like accuracy, reliability, fairness and security — not subjective experience. Fluent output is evidence a system predicts patterns well, not evidence anything is being felt or understood.
The framework Canada’s privacy regulators cite for evaluating a generative AI tool names seven characteristics to check it against: Valid and Reliable, Safe, Secure and Resilient, Accountable and Transparent, Explainable and Interpretable, Privacy-Enhanced, and Fair (NIST AI RMF, a U.S. standard, cited via the OPC’s own footnote to it).
Not one of the seven asks whether a system has an inner life. All seven ask whether its behaviour can be trusted, audited and controlled — a checklist for what a tool does, not for what, if anything, it experiences while doing it.
NIST’s own generative AI risk profile describes the underlying mechanism directly: these systems “generate outputs that approximate the statistical distribution of their training data,” predicting the next plausible token in a sequence. (NIST, Generative Artificial Intelligence Profile, U.S. framework)
That mechanism runs identically whether the resulting sentence turns out true or fabricated. A system built to produce a fluent, assured “I understand how you feel” is not thereby demonstrating that it feels anything — the sentence and the grocery list after it come from exactly the same next-token process, which is a fact about how the output is generated, not about what, if anything, accompanies generating it.
Not wrong, just outside what any current framework is built to answer. Nothing in Canadian AI governance regulates for or against machine consciousness, because none of it is designed to detect it — the working definitions in use describe a data-processing capability, not a claim about experience. See is AI actually intelligent for the closely related, but separate, question of how far that capability actually extends.
Start with how to evaluate a specific AI use before deciding where it fits.
