Canada Is Getting Interesting
This post is part of my Medium blog.
I've been paying attention to Canada lately, and it took me a while to admit why. I set the trilogy there, so I have the standing excuse. Rob Coleman flies out of Vancouver at five in the morning in Redundant, and by Book Three he's the chief commissioner of the office that exists to keep a human name on critical AI decisions. When I wrote that, it was fiction's convenience. Lately it reads like reporting.
Here's what surprised me when I dug into the actual law on both sides of the border. Neither country has the big statute. Canada's Artificial Intelligence and Data Act, the closest thing they had to a comprehensive AI law, died when Parliament was prorogued in January 2025. Congress hasn't passed anything either. On paper it's a tie. Two countries, no omnibus AI act between them.
One of them is governing anyway.
Canada's Treasury Board has run a Directive on Automated Decision-Making since 2019. It isn't a white paper. It binds every federal department that builds or buys a system to make or support an administrative decision: you run an algorithmic impact assessment first, you keep logs, you document, human review stays in the path, and higher-impact systems get looked at by other departments' reviewers before they're allowed near the public. It was updated in June 2025, and the deadline for older systems to come into compliance landed this past June. Because it attaches to procurement, it leaks into the private sector. If you want to sell decision systems to the Canadian government, your product has to live up to it.
Then there's the provincial layer, which is harder to wave off as Ottawa paperwork. As of January 1, 2026, an Ontario employer with 25 or more employees has to say so in the job posting if AI screens, assesses, or selects applicants. Not a guideline. An employment standard, enforced like any other, with a deliberately broad definition of AI so nobody lawyers their way around the chatbot. Quebec's Law 25 already obliges organizations to tell people when a decision about them was made exclusively by automated processing. The bank regulator published a model-risk guideline in September 2025 covering machine learning, effective 2027, and the banks answer to it. Even the Federal Court published principles for how it will and won't use AI in its own work.
None of that is dramatic. That's the point. Disclosure rules, impact assessments, audit trails, a regulator writing down what it expects from a bank's models. Nobody in Canada is claiming to have solved AI. They're doing what a functional government does with a general-purpose technology: start where it already touches people — hiring, credit, insurance, a decision about your file — and write the rule there.
The American side of the border, I mostly read with my mouth open.
The US federal government has no AI statute. The one standalone federal law that substantively regulates AI at all is the TAKE IT DOWN Act, which is about nonconsensual intimate imagery, signed in May 2025. Everything else is executive orders, and they point away from governing. In December 2025 the White House signed one that stands up a task force inside the Justice Department whose mandate is challenging state AI laws in federal court. In March 2026 the administration sent Congress a framework asking it to preempt the states outright. The FTC was directed to classify state-mandated bias mitigation as a deceptive trade practice. I read that sentence twice too.
The actual governing is happening in the states. Roughly thirty of them have AI law now. Colorado passed a high-risk AI act. California put frontier-model safety requirements and training-data disclosure into force on New Year's Day. Illinois requires annual independent audits of frontier models. Those are the only binding AI rules an American company answers to, and the flagship federal project in this space is a DOJ unit that exists to challenge them in court. The labs are even chipping in: xAI sued Colorado over its AI Act in April.
I write fiction about what happens when a country decides the machines are fine and the paperwork is the problem. In Essential, Parliament's answer to a catastrophe is a law that requires a certified human being, personally, in the loop for any critical system — with a four-year sunset, because even the people who wrote it weren't sure they were right. The American government in that book wants the cage protocol. The tech, none of the strings. I made that up. Then I watched a real administration look at its own states and call the strings the problem.
There is a version of the American position that's an argument. Existing law already covers this; let's enforce it before we build something new. Reasonable people hold it. That is not what's happening. What's happening is anti-regulation as an affirmative project: the federal government using its power to stop its own states from thinking about AI at all. And against that, the country that couldn't pass its big law is the one behaving like an adult. Impact assessments, disclosure, a human kept in the review path. Nobody gets a magazine profile for that work. It's still the work.
I keep watching Canada for the least cinematic reason a thriller writer can watch a country. They're doing the paperwork. We're suing whoever tries.