On 27 September 2026, Canada's AI Minister Evan Solomon told CTV's Question Period that AI deployment needs regulation. He offered no details. Asked whether Ottawa would legislate to slow or halt deployment, he said: "I'm not going to say 'yes, this is what we're doing.'" He called a just-completed federal consultation on AI transparency and safety "a very important first step" (CP24).
The caution is understandable. The result is that Canada has an AI minister, an AI strategy and a consultation, but no AI statute.
The strongest case for the minister's approach
The argument for patience is real. Frontier AI is changing fast, and a statute drafted in 2022 can look dated by the time it passes. The Artificial Intelligence and Data Act (AIDA) was criticised for leaving its core definitions, such as "high-impact system", to regulations written after passage. Starting with a consultation on narrower questions is sensible. Those questions include labelling AI content, disclosing chatbots, and reporting serious incidents. Solomon's line that "AI adoption moves at the speed of trust" also has merit. Disclosure rules are cheap for developers, useful to users, and do not restrict lawful speech.
What is actually on the table
AIDA was one of three enactments in Bill C-27, the Digital Charter Implementation Act, 2022. According to Parliament's LEGISinfo record, the bill was last in committee. That committee's final listed meeting was on 26 September 2024, and report stage, third reading and the Senate were never reached. The session ended on 6 January 2025, when Parliament was prorogued. The bill lapsed with it and has not been replaced.
What exists now is a set of separate items:
- a national AI strategy, which Solomon calls AI for All;
- the Protecting Privacy and Consumer Data Act;
- the Safe Social Media Act;
- announced or planned legislation on deepfakes, non-consensual sexual imagery, election protection and AI chatbots.
Solomon cited all of these on Question Period. None is a framework for AI systems as such.
The consultation he called a first step ran from 23 July to 23 September 2026. It covered five areas: identifying AI-generated content, telling people when they are dealing with an AI system, clearer information about system capabilities and limits, tracking serious incidents, and monitoring AI agents (Government of Canada). That is a reasonable list, but it is a list of questions. It commits the government to nothing.
The part that should worry innovators
The more consequential remarks were about slowing AI. Solomon said he had discussed slowing the release of AI models, reporting rogue-agent incidents, and a possible international technology stability board with G7 and G20 counterparts. No timeline has been set (CP24). In a later interview he described a framework to oversee "testing, model evaluation and model release" (The Hub).
This is where the absence of a statute matters. A body that approves or delays model releases would be a licensing regime in practice. Whatever it is called, it should be created by Parliament with defined triggers, criteria, appeal rights and sunset or review clauses. It should not be assembled through ministerial discussion and international side-channels. Release-gating also touches speech. Model outputs, open-weight releases and research publication are all expressive activity, and a discretionary delay power invites the problems the Electronic Frontier Foundation raised in a parallel context. Commenting on California's AI executive order, EFF warned that "government-controlled kill switches run the risk of being used as a form of retaliation against protected speech" (EFF). That statement concerns California, not Canada. The structural risk it names applies to any discretionary control over deployment.
There is also a tension with the government's own growth goals. Solomon has said he wants AI adoption to rise from 12 to 60 percent and projects $200 billion in GDP growth (The Hub). Uncertainty about whether releases could be slowed is itself a cost for Canadian firms deciding where to build and deploy.
What proportionate regulation would look like
The government can reconcile safety and innovation by doing three things.
- Legislate the narrow, evidence-backed pieces first. Chatbot disclosure, labelling of synthetic media in elections, and serious-incident reporting address documented harms. They can be drafted precisely and passed without waiting for an international body.
- Define any release-oversight power in statute. If a stability board or model-evaluation regime is coming, Parliament should see the text. It should cover who is in scope, what evidence triggers a delay, how long a delay can last, and who can challenge it. A vague power to "slow" releases is the opposite of proportionate.
- Publish what the consultation found. The government should release a summary of the responses and say which proposals it will pursue. Businesses cannot plan around a first step with no stated second.
Solomon is right that governments are accountable in ways AI companies are not, and right that rules can build trust. But trust comes from rules that are written down, debated and reviewable. Since AIDA lapsed with Bill C-27, Canadian developers and users have been told that regulation is coming, with no text to read. A consultation can inform a law, but it cannot replace one.