AI for All: The Missing Question of Distribution

Canada’s AI for All strategy sets out an important vision for expanding AI literacy, adoption, infrastructure, and access, while also emphasizing safety, trust, and protection from harm. One area that deserves greater attention is how AI is distributed: which capabilities are made available, to whom, and under what conditions. 

Broad access to AI’s benefits should remain a national priority. However, not every capability carries the same level of risk. As AI becomes more powerful and easier to use, access to its more consequential functions should reflect the harm they can enable. 

We already apply this principle to other beneficial tools. Medication can improve health and save lives, yet not every medicine is distributed in the same way. Some are available openly, some are kept behind the counter, and others require a prescription or professional supervision. Access depends on the level of risk, the knowledge needed for safe use, and the potential consequences of misuse. 

These safeguards do not diminish the benefits of medication. They help preserve those benefits while reducing foreseeable harm. AI is different in form, but the same principle can apply: broad public benefit does not require identical access to every model, function, or level of autonomy. 

When governance follows access 

This issue becomes more urgent when AI tools can enable serious harm at scale. Nudification tools, realistic deepfake generators, and emotionally persuasive companion chatbots have become accessible before consistent expectations were established for age restrictions, safety testing, or limits on higher-risk functions. In many cases, governance has followed only after the technology reached users and harm occurred. 

The creation of sexualized images of children shows the consequences of this reactive  approach. Generative image, video, and voice tools have valuable applications in accessibility, education, communication, entertainment, and creative work. The same technologies can also enable fraud, harassment, impersonation, manipulation, and sexual exploitation. 

In Thorn’s survey of 1,200 young people aged 13 to 20, one in 17 participants aged 13 to 17 reported having a deepfake nude created of them. In separate Thorn research, roughly one in ten minors surveyed reported knowing of cases where friends or classmates had used generative AI to create nude images of other children. 

These findings suggest that harmful functionality is no longer limited to technically  sophisticated users. It can be packaged into simple websites, applications, and subscription services requiring little or no coding knowledge.

The U.S. National Center for Missing and Exploited Children has reported sharp increases in CyberTipline reports involving generative AI and child sexual exploitation. The Internet Watch Foundation has also documented realistic AI-generated child sexual abuse material and subscription services that allowed users to generate customized imagery, including models apparently based on known victims. 

The problem therefore extends beyond individual misuse. Harmful functionality can be developed, packaged, commercialized, and distributed to people who may require almost no technical expertise. 

Similar concerns arise with emotionally persuasive chatbots. Companion-style features may benefit some users, but stronger safeguards may be needed when these systems are made available to children or other vulnerable groups. 

Moving safeguards upstream 

AI is not itself a moral actor. It does not understand ethics, experience harm, or accept responsibility. Responsibility belongs to the people and institutions that design, release, distribute, implement, and use it. 

This does not mean AI is neutral. Design and distribution decisions shape which actions become easier, cheaper, and more scalable. AI may not create malicious intent, but it can dramatically reduce the skill, time, and resources required to turn that intent into harm. Governance, therefore, cannot depend entirely on every end user choosing to act responsibly. 

Work at the implementation and use levels remains essential. Organizations must still consider how AI is deployed, how decisions are made, how users are affected, and how harmful outputs are detected and addressed. However, these measures should not carry the full burden of preventing harm. Where risks are foreseeable, governance should also intervene earlier by shaping how AI is released, which functions are made available, and who can access them. 

A proportionate distribution model could preserve broad access to ordinary, low-risk tools while introducing stronger safeguards as risks increase. Depending on the context, these safeguards could include age-appropriate or verified access, restrictions on specific high-risk functions, staged releases, monitoring, user education, limits on onward redistribution, or stronger conditions for downloadable and modifiable models.

This would not require restricting AI broadly or creating unnecessary barriers to innovation. It would recognize that access is not a simple yes-or-no decision. The same technology can be provided through a monitored consumer service, a controlled application interface, limited research access, or unrestricted downloadable files. Each method transfers a different degree of control and creates different opportunities for misuse.

Distribution-level governance would complement responsible design, implementation, content moderation, and enforcement by moving some responsibility earlier in the lifecycle, before harmful functions become easy to access, modify, commercialize, and redistribute. 

This approach could reduce preventable harm, clarify responsibility across the AI supply chain, and lessen the burden placed on victims to report material after it has already been created and circulated. It could also strengthen public confidence by showing that wider AI adoption does not require treating every form of access as equally safe. 

AI for All sets an important national direction, but literacy, adoption, and infrastructure are only part of responsible access. Canada should make distribution-level governance an explicit part of implementing the strategy. 

Where serious risks can be anticipated, safeguards should shape how AI is released, who can access it, and what responsibilities accompany that access. 

Canada’s vision of AI for All will be stronger if wider access is matched by greater care in how higher-risk AI is released and distributed. 

References 

Government of Canada. Canada’s National Artificial Intelligence Strategy: AI for All. 

Thorn. Deepfake Nudes and Young People. 

Thorn. Report: 1 in 10 Minors Say Peers Have Used AI to Generate Nudes of Other Kids. 

National Center for Missing and Exploited Children. Generative AI. 

Internet Watch Foundation. AI-Generated Images, Annual Data and Insights Report 2025.