Canada’s AI Strategy Pushes Adoption Before Clear Rules
Matthew Leo · Published July 22, 2026 · AI
Canada Wants an AI Boom. The Rules Can Apparently Come Later.
Canada has a new plan for artificial intelligence, and it comes with some very confident numbers.
Raise business adoption from about 12% to 60% by 2034.
Create up to 250,000 jobs connected to AI adoption by 2031.
Provide 90,000 AI-related jobs and work placements for young Canadians.
Unlock nearly $200 billion in economic growth.
Build a public supercomputer, expand Canadian data centres and help domestic AI companies become global players.
The government calls the strategy AI for All.
It is ambitious. Parts of it are smart. Canada needs stronger technology companies, better infrastructure and less dependence on foreign platforms.
But there is something uncomfortable about how the plan is written.
The targets for adoption, investment and growth are specific.
Many of the protections are not.
Canada knows how many businesses it wants using AI. It is less clear about what those businesses must tell workers, customers and the public when AI is used to make decisions about them.
The country wants the boom first.
The rulebook is still being assembled.
Canada Does Need an AI Strategy
Criticising the plan does not mean Canada should sit out the AI race.
That would be ridiculous.
Canada has internationally respected researchers, strong institutions such as Mila, Amii and the Vector Institute, and thousands of companies already developing AI products. The federal strategy says more than 150,000 Canadian jobs are directly associated with AI across over 3,500 companies.
The problem is that Canada has often been better at producing research than turning that research into large Canadian companies.
The strategy says nearly 70% of Canadian-led startups eventually become headquartered outside the country. Canadian firms also rely heavily on foreign cloud providers for the computing power needed to build and run their products.
That dependence is not harmless.
A Canadian hospital, government department or company may be using an AI system built abroad, hosted abroad and governed partly by another country’s laws. Foreign providers can change their prices, restrict access or remove products that Canadian organizations have built their work around.
Prime Minister Mark Carney has warned that foreign AI platforms could give outside governments and companies access to Canadian data and influence systems that shape Canadian lives.
Building more Canadian capacity is therefore sensible.
The question is not whether Canada should support AI.
It is whether support should arrive faster than accountability.
The Economic Promises Are Enormous
AI for All expects business adoption to rise from roughly 12% today to 60% by 2034.
It projects up to 250,000 new jobs through AI adoption by 2031 and a 3% increase in gross domestic product, worth close to $200 billion.
Those figures are projections, not cheques waiting to be deposited.
AI could increase productivity. It may help doctors process information, help farmers use fertilizer more carefully and help small companies complete work that once required larger teams.
But productivity forecasts have a habit of arriving before the evidence.
A company buying an AI tool does not automatically become more productive. It may simply add another subscription, another security risk and another piece of software employees have been told to use.
Some businesses will save time.
Some will spend months paying consultants to discover that their real problem was a badly organized spreadsheet.
Canada’s adoption target also assumes that more AI use is automatically desirable. The government describes the low adoption rate as a gap that needs to be closed.
Sometimes a company has not adopted AI because it lacks knowledge or resources.
Sometimes it has not adopted AI because the technology does not solve an important problem.
The strategy itself notes that 78% of non-adopting firms say they do not see how AI benefits their products or services.
That is not always a failure of education.
Sometimes the sensible answer to “Where should we put AI?” is “Nowhere yet.”
The Government Wants to Become an AI Customer
Canada plans to use the federal government as an early buyer of Canadian AI products.
The strategy says Ottawa will become a strategic “anchor customer,” use Buy Canadian policies and speed up the purchase and deployment of AI across the public service.
That could help Canadian companies.
Landing a government contract gives a young business revenue, credibility and a customer large enough to test its product properly.
It could also create pressure to deploy systems before departments understand them.
Governments do not merely use software to sell shoes or recommend films. They process immigration files, benefits, taxes, health information and other decisions that can change a person’s life.
AI for All says meaningful public-sector decisions will retain human oversight. It also promises high standards for transparency, privacy and accountability.
Those commitments are necessary.
They are still broad.
“Human in the loop” can mean a trained public servant carefully reviewing an AI recommendation.
It can also mean an exhausted worker clicking approve because the system handles thousands of cases and disagreeing requires extra paperwork.
Human oversight is useful only when the human has enough time, information and authority to challenge the machine.
Canada Still Has No General AI Law
Canada previously tried to introduce a broad regulatory framework through the Artificial Intelligence and Data Act, commonly called AIDA.
That legislation was part of Bill C-27. It died when Parliament was prorogued in January 2025. As of July 2026, Canada still has no single law creating a general framework for AI systems across the private sector. Existing privacy, human-rights, consumer-protection and other laws continue to apply, but they form a patchwork rather than one clear AI regime.
There were valid criticisms of AIDA.
Its definitions were vague. Civil-liberties and human-rights groups argued that it gave too much control to the department responsible for promoting the AI industry. Others believed copying the European Union’s approach would create excessive bureaucracy.
Letting a flawed bill die was not necessarily a mistake.
Failing to replace it with clear obligations would be.
AI for All promises updated privacy laws, online-safety rules, model testing, watermarking and a Trusted AI Certification program. It also adds $50 million to expand the Canadian AI Safety Institute.
Those are useful pieces.
They are not a complete answer to how AI should be used in hiring, insurance, credit, policing, education or the workplace.
Canada is choosing targeted rules instead of one large AI act.
That approach can work.
It still needs actual rules.
To Be Fair, Some Rules Have Started Arriving
The government did not publish the strategy and then completely disappear into a cloud of innovation slogans.
On June 15, it introduced Bill C-36, which would create the Protecting Privacy and Consumer Data Act.
The proposed law would recognize privacy as a fundamental right, strengthen protections for children, require more transparency around automated decisions and allow significant penalties for serious violations.
Canada also introduced the Safe Social Media Act, which would place new duties on social media companies and certain AI chatbot services, particularly where children and harmful content are concerned.
These are meaningful proposals.
Neither bill is the same as a general AI law, and neither has completed the parliamentary process.
Privacy legislation may control how personal information is collected and used. Online-safety legislation may address harmful chatbot behaviour and dangerous content.
They do not automatically answer questions such as:
When must an employer tell workers that AI is measuring them?
Should a person have the right to appeal an AI-assisted hiring decision?
What must a company disclose about the data used to train a model?
When can police use facial recognition or predictive systems?
Who is responsible when an AI agent takes an unauthorized action?
Those questions should not be left entirely to voluntary codes and future court cases.
Workers Are Promised a Voice, Not Given One
AI for All repeatedly describes its approach as “pro-worker.”
It says workers and unions should help shape workplace adoption. It promises training for young people, mid-career workers, tradespeople and workers in sectors likely to face early disruption.
That is better than pretending AI will create only winners.
The strategy even says productivity gains should flow into paycheques rather than only into profits.
Excellent sentence.
Now turn it into policy.
The strategy does not itself give workers a legal right to be consulted before AI is introduced.
It does not require employers to provide notice when automation may remove jobs.
It does not guarantee that time saved through AI becomes shorter hours, higher wages or less stressful work.
Without rules, “pro-worker AI” depends heavily on the goodwill of the employer.
A company can use AI to remove repetitive paperwork and give staff more time for meaningful work.
It can also cut two positions, keep the workload unchanged and describe the result as efficiency.
Both outcomes count as adoption.
Only one helps workers.
Job Creation Is Only Half the Story
The government projects that AI adoption will create up to 250,000 jobs by 2031.
That sounds reassuring, but a national total can hide painful changes underneath it.
AI may create software, infrastructure, sales, training and cybersecurity jobs while reducing demand for some administrative, creative and entry-level roles.
The new job may be in a different city, require different qualifications or appear three years after the old one disappeared.
Telling a laid-off customer-service worker that AI created a data-centre engineering position somewhere else does not solve the worker’s problem.
Training matters, but “learn new skills” has become the standard answer governments give whenever technology threatens someone’s livelihood.
Not everyone can become an AI engineer.
Nor should they have to.
A serious worker policy would track which jobs are being removed, require advance notice for major AI-driven changes and ensure that workers share in productivity gains.
The strategy promises measurement, workforce alliances and cooperation with unions.
Those are good beginnings.
They are not protections.
Copyright Somehow Missed the Strategy
Canada’s AI strategy is about 50 pages long.
The word copyright does not appear once.
That is a remarkable omission for a plan about a technology built partly by consuming enormous amounts of writing, art, music, journalism and other human work.
Canadian publishers are already suing OpenAI, alleging that their journalism was used to train ChatGPT without consent or compensation. Artists and writers have been asking whether AI companies should be allowed to train models on copyrighted work without permission.
The strategy promises a $50 million Creative Technology Program to help Canadian creators use AI on their own terms.
Helping creators use AI is not the same as deciding whether AI companies may use creators.
That is the missing half.
The government has said more copyright work is coming and that creators will not be forgotten.
“Stay tuned” is not much comfort when companies are already training and selling models today.
Canada does not need to ban AI training.
It does need a clear position on consent, compensation, credit and the disclosure of training material.
Creators should not have to sue their way into national policy.
Transparency Is Still Too Voluntary
The government wants watermarking for AI-generated content and a certification program for trustworthy products.
Both could help.
Neither tells Canadians enough about what happens behind the screen.
Technology-law professor Michael Geist has argued that the strategy should require disclosure of where AI is being used, what works were included in training data and what information governments or law-enforcement agencies request from AI companies. Instead, much of the current transparency plan is left to future standards and voluntary certification.
A trustworthy logo is useful only if the standards behind it are strict, independent and regularly tested.
Otherwise, it becomes the AI version of a green leaf printed on a plastic bottle.
Canada should require clear public information when high-impact AI is used.
Who built the system?
What is it deciding?
What data does it use?
How accurate is it across different groups?
Can a person challenge the result?
Has the system caused previous incidents?
Those are not trade secrets.
They are the minimum information needed for accountability.
Sovereign AI Is a Good Goal With an Expensive Bill
One of the strongest parts of the strategy is its focus on Canadian infrastructure.
Canada plans to build a public supercomputer, provide an additional $700 million in affordable computing access for small and medium-sized companies, and support large AI data centres. Proposed partnerships could provide 850 megawatts of capacity by 2030, with room to scale as high as 2.3 gigawatts.
Canada also plans a $500 million technology growth fund and may take ownership stakes in promising Canadian AI companies.
There is a real argument for this.
Without domestic computing power, Canadian firms send money, data and intellectual property to foreign cloud companies. They may build Canadian products on infrastructure Canada cannot control.
But sovereignty is not achieved merely by placing servers on Canadian soil.
Who owns the data centre?
Who owns the chips?
Who controls the software?
Can a foreign court order the provider to hand over Canadian data?
Will public money build infrastructure whose profits ultimately flow abroad?
The word “Canadian” needs to describe control, not only geography.
The Environmental Promises Need Numbers Too
AI data centres use enormous amounts of electricity and sometimes large amounts of water for cooling.
The strategy acknowledges this. It says Canada’s cleaner electricity grid and colder climate could make the country an attractive place to build lower-emission data centres. It promises to connect new development with clean-energy expansion, environmental standards and benefits for local communities.
That is sensible.
The environmental commitments should be as measurable as the economic ones.
Canada has a 60% business-adoption target.
Where is the national target for data-centre water efficiency?
Where are the public reporting rules for electricity use?
What must a company contribute when its data centre requires new transmission lines or generation capacity?
How will communities be consulted before a large project competes for local water and power?
Canada plans to double its electricity infrastructure by 2050 partly to meet demand from industry, electric vehicles and AI data centres.
Taxpayers and ratepayers should know how much of that expansion supports public needs and how much subsidizes private computing.
A data centre is not automatically a nation-building project because someone attached the word AI to it.
The Strategy Gets Open Source Right
AI for All deserves credit for supporting open-source AI.
The strategy argues that relying on a small number of closed foreign platforms creates vendor lock-in, high costs and limited control. Open tools can be cheaper, easier to adapt and better suited to private or on-premises use.
This may be one of Canada’s smartest opportunities.
Canada is unlikely to outspend the largest American and Chinese companies in a race to build the biggest model.
It may be able to help create a trusted ecosystem of smaller, useful and adaptable systems for health, government, Indigenous languages, agriculture and Canadian businesses.
That would be more valuable than spending billions to announce a Canadian chatbot that arrives three years late and answers questions with a maple leaf beside the text box.
Canada should compete where it has an advantage, not where the biggest company has already won through scale.
Rules Do Not Have to Stop Innovation
The technology industry often presents regulation as a choice between progress and paralysis.
That is convenient for companies that prefer making their own rules.
Bad regulation can absolutely slow innovation. Vague laws can create uncertainty, and excessive requirements may hurt small companies more than global giants with entire compliance departments.
But no rules create uncertainty too.
A business may hesitate to adopt AI because it does not know whether using customer data will later become illegal.
A hospital may fear buying a system whose liability is unclear.
Workers may resist technology they believe will be used secretly against them.
Creators may fight every AI proposal because they expect their work to be taken without compensation.
Clear rules can build trust and make responsible adoption easier.
The government’s own strategy says trust is necessary for adoption.
It should follow its own logic.
Canada Should Decide What Success Means
The AI for All strategy treats adoption, jobs and economic growth as major measures of success.
Those numbers matter.
They are not enough.
Canada should also measure:
How many workers received higher pay or shorter hours because of AI productivity?
How many people successfully challenged an automated decision?
How many AI-related privacy breaches occurred?
How much public money went to Canadian-controlled companies?
How much electricity and water did AI infrastructure consume?
How many creators gave consent or received compensation for training data?
How often did public-sector AI systems produce incorrect or discriminatory outcomes?
A country can reach 60% adoption and still build an AI economy that most people resent.
Technology spreading is not the same as technology succeeding.
Build the Boom and the Guardrails Together
Canada is right to worry about falling behind.
It is right to build infrastructure, support Canadian businesses, train workers and reduce dependence on foreign technology.
There is plenty to like in AI for All.
The problem is the order of confidence.
The government speaks with certainty about growth, jobs, adoption and investment.
On copyright, worker rights, transparency and broader AI accountability, Canadians are frequently told that consultations, standards or legislation will follow.
By the time the rules arrive, AI may already be embedded in workplaces, classrooms, government services and hiring systems.
Removing a harmful system after everyone depends on it is much harder than setting conditions before it spreads.
Canada does not need to choose between becoming an AI leader and protecting Canadians.
A country that can find billions for computing power can find the courage to write clear rules.
The boom and the guardrails should be built at the same time.
Otherwise, “AI for All” may end up meaning that everyone experiences the technology while only a few decide how it works, who profits and who pays when it goes wrong.