TORONTO — The federal government’s AI advisers are divided on how—or even if—Canada should create new rules around the technology’s development and use. Some civil society groups, meanwhile, say Ottawa isn’t asking the right people or questions as it considers regulating AI. 

In memos submitted late last year, members of the AI Strategy Task Force called for new security plans as AI is built into more economic and military applications. On the regulatory front, they proposed a range of measures, including the introduction of strict laws and limitations on the technology to advance digital safety, binding requirements for AI developers, or voluntary standards and programs to encourage them to act ethically. 

Some task force members claimed now is the wrong time for Canada to impose sweeping AI regulations. Benjamin Bergen, then-president of scaleup lobby group Council of Canadian Innovators, said it was too late for Ottawa’s rulemaking to shape the development of the most advanced models, and too early to try to address future applications that might emerge. He added that a long legislative and regulatory process risks “introducing considerable legal uncertainty at a pivotal moment for the evolution of AI technologies.”

The Liberal government tried and failed to pass an AI law in the last Parliament. AI Minister Evan Solomon—who appointed the task force to advise on the upcoming renewal of the national AI strategy—has suggested he won’t revive the legislation wholesale. Still, he’s said that widespread AI adoption will depend on getting the public to trust the technology.

In his submission, task force member and Build Canada chair Daniel Debow said Solomon shouldn’t revive the AI law and should also “avoid new regulations” which could “stifle AI adoption.” Debow called for Ottawa to instead apply existing privacy, security and other laws to AI. If the federal government must regulate, Debow said, its rules should be “clear and lightweight,” and ready within six months. 

Cohere chief AI officer Joelle Pineau also recommended Ottawa focus on where AI fits into existing laws. The federal government should set up a new unit to coordinate AI policy across departments and help them set specific rules for the sectors they oversee, she said. The U.K. has similarly delegated AI rulemaking to industry regulators. Ottawa established a central AI secretariat in July 2024, but it moved to the Innovation Department to support Solomon after he was appointed to cabinet last year. 

To address AI’s trust and safety issues, the country first needs a common understanding of how and when the technology can go wrong, according to Pineau. She proposed a “taxonomy of risks” which would focus on those that can be seen, measured and cause material harm to people or organizations. For example, an AI tool could cause physical hazards like encouraging self-harm or non-physical ones like defaming someone or violating their IP rights. Ottawa could also set out guidelines for how AI developers and users should evaluate and report those risks, Pineau said.

Some task force members recommended voluntary measures and incentives. Coveo executive chair Louis Têtu called for national standards that are “secure, scalable and repeatable.” Bergen said Canadian firms should also be involved when international bodies are setting standards. In separate memos, University of Waterloo engineering dean Mary Wells and former Canadian Tire executive Cari Covent suggested Ottawa use its own buying power, tweaking the contracting process to favour firms that meet ethics and safety criteria. “Government procurement is a powerful lever to shape and ensure responsible AI development,” Wells wrote.

Ottawa could also launch a “Canadian AI trust label” for generative tools that are secure, explain themselves and leave humans in control, said Olivier Blais, co-founder of consultancy Moov AI. Firms that got the seal of approval would gain a reputational lift, he claimed. All of the submissions made by the task force can be viewed here. 

Some task force members called for more binding requirements for those building and using AI. Ottawa should mandate that developers, policymakers and public servants get “ethics and bias training,” Wells said, while firms should be required to maintain codes of practice and audit documentation.

Shelly Bruce, the former chief of the Communications Security Establishment, said the federal government should set minimum security terms for firms running large or widely used AI models. That includes setting up cyber defences, commissioning independent assessments and recording all the components that go into their systems. Bruce also said developers should be required to “report critical incidents and near-misses.” Covent, meanwhile, recommended Ottawa establish a national system to track and investigate AI mishaps beyond security, such as self-driving cars getting into accidents or AI-driven services making major errors. 

Many task force members said Ottawa needs to be transparent about its own use of AI, including via a public registry of its automated tools. “By giving visibility into what and how AI is used in government, we can increase transparency and trust in AI usage across the population,” Pineau wrote. The Treasury Board Secretariat recently launched such a registry, although it’s currently just a spreadsheet.  

While the Liberal government’s proposed law targeted systems that could have a high impact because they were used in sensitive settings, tech firms have argued that the rules should instead focus on the riskiness of the applications.

Other task force members said it’s time to legislate. Canada should establish “clear prohibitions on specific AI uses that pose unacceptable risks,” like biometric surveillance in real time or systems that judge guilt, said Doyin Adeyemi, a law and management student at the University of Toronto. The EU AI Act includes similar bans on certain applications of the technology.

Several memos called for specific protections for children interacting with AI. Digital platforms should be required to conduct and share assessments of the risk or impact to minors from using them, and remove harmful content that involves them, said Adeyemi and Taylor Owen, founding director of a tech-focused thinktank at McGill University, in separate submissions. 

Adeyemi also called for Canada to adopt California-style rules requiring chatbots to remind children that they’re chatting with an AI, not a human, and to have protocols to handle mental health incidents. That would help address “unique risk of emotional dependency and parasocial bonding between minors and AI companions,” she said. 

Canada needs new regulators to deal with AI and digital platforms, according to some submissions. Adeyemi recommended an AI equity commission bringing together people with legal, civil rights and technical backgrounds with members of Black, Indigenous and other racialized communities. The regulator would have the power to conduct audits and monitor compliance with the no-go rules. 

Owen—who previously advised the Liberal government on what would eventually become its online harms policy—suggested a new digital safety commission. The agency would regulate social platforms and consumer-facing AI applications, auditing their algorithms and issuing orders or penalties when they break the rules. 

Canada needs to be prepared for the security risks that AI generates, several task force members said. Bruce recommended Ottawa tag any technology that’s deployed at the national level or that’s widely used by civilians as a “system of importance,” which lets the Communications Security Establishment give firms running such software advice and technical support. Energy and data infrastructure are becoming “increasingly important military targets,” and Canada also needs a security plan to protect them, said Sam Ramadori, co-president of LawZero, a research non-profit. 

Some civil society and digital rights advocates say the Liberal government’s process for renewing the national AI strategy is dominated by industry interests, and want a redo. Its consultation disregards “the Canadian public’s known and wide-ranging concerns about the demonstrated risks and harms” of the technology, according to an open letter published in October. It said AI can further erode labour rights, automate racism, intensify gender-based violence, marginalize LGBTQ2SIA+ people, exclude people with disabilities and enable mass surveillance. 

Initial signatories included the British Columbia Civil Liberties Association, OpenMedia, as well as staff from the surveillance-tracking Citizen Lab and dozens of academics. Some backers have since launched their own “people’s consultation” to feed a competing national AI strategy. 

While Solomon won’t revive the AI law as is, he has promised to shortly introduce new privacy legislation. Last month, he said the updated AI strategy will come out this quarter.