Canada's Minister of AI and Digital Innovation, Evan Solomon, opened a public consultation on July 23, 2026 asking Canadians how the federal government should regulate transparency in AI systems. The consultation runs through September 23, 2026 and is built around a discussion paper, "Enhancing trust in artificial intelligence through increased transparency," published by Innovation, Science and Economic Development Canada (ISED). It asks pointed questions across five areas: how to detect and label AI-generated content, when disclosure is owed to someone interacting with an AI system, what information the public should get about a system's capabilities and limits, how serious AI incidents should be defined and reported, and — the newest addition — who is responsible for disclosing that an autonomous AI agent, rather than a human, took an action on a user's behalf.
The scope is the story. This is not a revival of the Artificial Intelligence and Data Act (AIDA), the comprehensive risk-tiered regime that was folded into Bill C-27 and that died on the order paper when Parliament was prorogued on January 6, 2025 following Justin Trudeau's resignation. AIDA would have imposed binding obligations — risk assessment, mitigation plans, recordkeeping, and potential criminal penalties — on developers and deployers of "high-impact" AI systems, defined through a schedule Ottawa never finished drafting. Solomon has been explicit that the old bill is not returning intact. As he put it to BetaKit in mid-2025, his approach is "light, tight, right" — light enough not to smother adoption, tight enough to close real gaps, sized to Canada's economy rather than borrowed wholesale from Brussels or Washington.
The case for moving faster
The strongest argument against Solomon's incrementalism is timing. Canada has now gone 19 months since AIDA's death with no binding AI-specific rules on the books, relying on PIPEDA, a privacy statute drafted in 2000, to catch whatever AI harms it can reach. Deepfake fraud, election-adjacent synthetic media, and opaque AI agents making purchases or filing claims on a user's behalf are not hypothetical risks reserved for some future framework — they are live problems today, and a non-binding consultation with no legislative deadline attached does not stop any of them. Privacy Commissioner Philippe Dufresne has publicly called for stronger enforcement powers to match international peers, and AI researcher Yoshua Bengio has argued Canada needs treaty-level coordination, not another comment period. The EU's comparison point lands squarely in the middle of this consultation window: Article 50 of the EU AI Act, which mandates labeling of AI-generated content, disclosure when a user is talking to a chatbot, and marking of deepfakes, entered into force on August 2, 2026 — a hard legal deadline that arrived three weeks after Ottawa merely started asking questions about the same territory.
Why narrow is still the right call
That comparison, though, cuts both ways. The EU AI Act's transparency provisions are binding precisely because they were carved out of a much larger, multi-year risk-classification apparatus that took European co-legislators three years to negotiate and that businesses are still struggling to interpret — the European Commission needed until July 20, 2026 to finalize guidelines on what Article 50 actually requires, a full two years after the Act's initial adoption. AIDA tried to import that scale of ambition into a single Canadian bill without the institutional runway to match, and it collapsed under its own complexity before a single high-impact system schedule was finalized. Rebuilding a comprehensive AIDA-style regime from scratch, in a minority Parliament where Solomon himself has noted legislative math is unpredictable, risks repeating that failure rather than learning from it.
A transparency-first approach — content labeling, incident reporting, agent disclosure — is also the part of AI regulation with the clearest evidence base and the least contested rationale. Disclosure rules are calibrated interventions: they inform without dictating design choices, they are enforceable through existing consumer-protection and privacy machinery rather than a wholly new compliance bureaucracy, and they let Ottawa observe how the EU's binding version performs in practice before deciding whether Canada needs equivalent teeth. Canada's June 2026 National AI Strategy, "AI for All," already commits to watermarking AI-generated content and a voluntary Canada Trusted AI Certification program — this consultation is the mechanism for turning those commitments into rules grounded in actual stakeholder input rather than a minister's guess.
The real risk is not that Ottawa is starting narrow — it is that narrow becomes permanent by drift. Solomon told reporters in October 2025 that legislation was coming, then let the timeline slip past his own original end-of-2025 target into this consultation, which itself has no statutory deadline for translating input into a bill. A proportionate, transparency-anchored framework is the right sequencing after AIDA's collapse. It stops being proportionate the moment it becomes Ottawa's permanent substitute for ever finishing the job.