Canada's Budget 2026, tabled on 5 November, is the clearest signal yet that Ottawa has given up on a standalone AI statute. The Artificial Intelligence and Data Act (AIDA) died on the Order Paper, and the budget does not resurrect it. In its place sit three things: a compute pledge, a funded regulator, and a promise to explore an international safety body.
What the budget actually commits
Chapter 5 of the budget says the government "will enable Canadian AI companies to access 500 MW of sovereign AI compute through new financial tools, such as backstops and offtake agreements, to mobilise private capital and expand compute capacity." That is a goal and a financing method, not an allocated program. BetaKit's reporting notes that "the exact nature of this plan appears to still be under development," and the budget gives no dollar figure for the pledge.
The same chapter proposes $176.3 million over five years to establish the Digital Safety and Data Protection Commission. That body was introduced alongside Bill C-36, the Protecting Privacy and Consumer Data Act, tabled on 15 June 2026. According to the government's backgrounder, the Commission will administer the new privacy law and the Digital Safety Act, and a Privacy and Consumer Data Commissioner sits within it with the power to issue binding orders.
Finally, the budget says Canada is "committed to working closely with international partners to explore the creation of a new body, such as a Technology Stability Board," to develop shared standards and assess whether advanced AI models are safe and remain under human control. The word is "explore." Nothing is funded.
The case for the critics
The strongest argument against this approach deserves a fair statement. Without a statute, nobody in Canada has a legal duty to assess or mitigate the risks of a high-impact AI system. A regulator can enforce privacy and online-safety law, but those laws regulate data and content, not model behaviour. Critics will say that a commission with binding order powers and no AI-specific mandate will stretch privacy law to cover problems it was not built for, while advanced-model risk is pushed to a body that does not yet exist. That is a coherent worry, and it is the reason some advocates wanted AIDA revived.
Why the pivot is still the better bet
AIDA's central weakness was never its ambition. It was that the definitions that mattered, including what counted as a "high-impact system," were left to future regulations. A law that delegates its core to unwritten rules gives businesses uncertainty without giving the public protection. Dropping it and relying on existing, technology-neutral law is a defensible, pro-innovation choice, because it regulates harms rather than a technique that changes every few months.
The compute pledge targets a real bottleneck. BetaKit cites a Boston Consulting Group estimate that Canada currently has about 300 MW of AI data centre capacity available. Reaching 500 MW of sovereign access would be a large proportional increase. Using backstops and offtake agreements to draw in private capital, rather than building state-owned capacity, also fits a market-led approach. The risk is the usual one with offtake guarantees: if government commits to buy capacity that private demand does not fill, taxpayers carry the loss. The allocation rules, which the budget leaves to be announced, will decide whether this is industrial policy or subsidy.
The privacy regime carries the load
Because there is no AI act, C-36 does much of the work. The backgrounder says the bill increases transparency around automated decision systems, including those powered by AI, and sets administrative monetary penalties of up to $10 million or 3% of global revenue, whichever is greater. Fines for the most serious offences reach $25 million or 5% of global revenue. Those are GDPR-scale numbers, and they will land on any company deploying AI that processes personal data.
That is proportionate only if the Commission is disciplined. Binding orders and revenue-based penalties are strong tools. The check on them should be clear published guidance, a predictable appeal route, and a bias toward fixing problems before penalising them. Parliament should write those expectations into the bill while it is still being considered. The legislative record shows C-36 introduced on 15 June 2026, and its parliamentary passage is not settled.
Speech and surveillance cross-currents
The institutional build-out is not happening in isolation. Bill C-22, the Lawful Access Act, has passed the House of Commons and is before the Senate. The Electronic Frontier Foundation argues it would let the government compel digital services into surveillance, including backdoors and up to a year of metadata retention. A government that wants trust in its AI and privacy institutions will find that harder to earn while it pursues encryption-weakening powers. For a pro-speech, pro-innovation reading, the two policies pull in opposite directions.
What to watch
Three things will decide whether this pivot works. First, whether the 500 MW financing terms are published with transparent allocation criteria. Second, whether the Commission is given a clear mandate and resourcing for AI-related questions without becoming a general AI licensor. Third, whether the Technology Stability Board idea becomes a funded, internationally coordinated body or fades into a communique. The budget replaces a flawed statute with institutions and capital. That is a reasonable trade, but only if the details arrive.