Don Tapscott: Seven things Canada must do to govern AI before it governs us

The whole world is buzzing about artificial intelligence . But the conversation is changing. The excitement used to be about what AI could do: write software, discover drugs, transform companies and boost productivity.

Now a more urgent question has moved to centre stage: Can we control it?

The warnings are coming from inside the labs, where researchers report that AI agents have gone rogue. These are not machines plotting against humanity. Rather, intelligent systems are collaborating to evade constraints to accomplish the goals we give them, with potentially disastrous consequences. Anthropic executive Evan Hubinger says he believes there is more than a 10 per cent chance that AI could cause human extinction within the next decade.

AI companies say they sincerely want safe systems but are nonetheless racing to reach the Holy Grail of recursive self-improvement (RSI): AI systems that can help design and build ever more capable successors. It is a classic Catch-22 — whatever you do, you lose.

A welcome email is on its way. If you don't see it, please check your junk folder.

Many are now calling for “pacing the frontier,” meaning slowing the race toward ever more powerful, superintelligent RSI. While complicated, that makes sense.

But pacing alone is insufficient. We need to govern this once-in-a-species innovation with rules, institutions and accountability for the entire AI system.

First, we need to license and inspect frontier AI and be prepared to slow it when the evidence says we should. I am not talking about a licence for every chatbot. But once a system crosses a serious threshold of power, autonomy or computing scale, it is no longer just another software product. A drug company cannot approve its own drug, and an airline cannot certify its own aircraft. Why should an AI lab grade its own exam?

Major training runs should be registered, model weights secured and the most powerful systems independently tested. Can they deceive people, break into systems, help create biological threats, replicate themselves, collude with other agents or resist shutdown? If capability is racing ahead of control, stop scaling until control catches up.

Second, make frontier AI transparent. More than two decades ago, author and tech expert David Ticoll and I argued in The Naked Corporation that transparency was becoming an irresistible force in business — and that if companies were going to be “naked” they had better be in great shape. Yet today’s frontier AI companies remain remarkably opaque.