Article 50 transparency obligations have applied since 2 August 2026. On the same date, the European AI Office and national authorities took up their supervision and enforcement powers. In Canada, the countdown to 1 May 2027 drops below nine months.
2 August 2026 changed what the AI Act is for Canadian organizations touching the European market. Until then the regulation was a calendar. Since then it is a regime with an authority, a procedure and fines.
One increasingly common reading needs resisting: the high-risk deferral does not empty 2 August of its content. Transparency, literacy and enforcement all arrived on schedule. Only Annex III moved.
Since 2 August 2026, providers of systems intended to interact directly with people must tell those people they are interacting with an AI. Systems generating synthetic content, whether audio, image, video or text, must make that content identifiable as AI-generated or AI-manipulated.
Deployers of emotion recognition systems, biometric categorisation systems, or systems generating deep fakes are also subject to disclosure duties.
These duties are demonstrated, not declared. A regulator asking the question wants the evidence: where the notice appears, at what point in the journey, on which systems, and since what date. That is system-by-system documentation work, not a policy update.
Obligations on general-purpose AI models had applied since 2 August 2025, but without an enforcement mechanism. Since 2 August 2026 the Commission can investigate and sanction. The European AI Office can request technical documentation, evaluate models, require corrective measures and impose fines.
General-purpose models placed on the market before 2 August 2025 have additional time and must comply by 2 August 2027.
If you embed a third-party general-purpose model, the question to put to your vendor has changed. It is no longer "are you compliant" but "what technical documentation can you produce, and how fast". That answer belongs in your evidence file, not only in theirs.
The AMF guideline on the use of artificial intelligence takes effect on 1 May 2027. OSFI Guideline E-23 on model risk management takes effect the same day. A federally chartered institution operating in Québec answers to both.
The AMF guideline sets out its expectations across seven elements of an AI system's life cycle, from the rationale for using it through to monitoring. E-23 covers all models that carry risk to the institution, AI and machine learning included, and extends to foreign bank branches and insurers.
Nine months is a governance cycle, not a documentation project. Inventorying systems, assigning risk ratings, validating models and producing evidence takes longer than a quarter. Organizations starting in the autumn of 2026 will arrive ready. Those starting in 2027 will arrive late.