Canada’s privacy regulator says people need more information about AI, including when AI uses personal information or helps shape decisions, and when AI produces harmful content about them. The Office of the Privacy Commissioner of Canada (OPC) raised those concerns in comments submitted to a government consultation, and the OPC also pointed to AI agents and the failures and limits of AI systems. For companies buying and selling AI tools, the OPC submission raises a question about how much people need to know to judge whether an AI tool can do the job it is intended to do.
AI Transparency Also Matters to the People Content Depicts
The consultation comes after ISED released a discussion paper. ISED is the Canadian department whose responsibilities cover innovation, science and economic development, and the paper names five areas where transparency matters: AI-produced material, exchanges with AI, information on AI systems, incidents involving AI and AI agents.
OPC praised the government for paying attention to transparency. OPC agreed that more openness could help build trust among the public and give consumers better information, and it could help with research and experimentation while making competition fairer for organizations. The consultation identifies these as possible benefits. The submission does not establish that any of those benefits have happened.
On AI-produced material, OPC questioned whether efforts to improve transparency give enough attention to everyone involved. Someone seeing an image online or reading other material can better decide what to do if they know AI produced it. But concentrating on that viewer could mean overlooking someone depicted or discussed in the content, and OPC suggested privacy authorities might have a part to play in ensuring transparency around this AI use. OPC also cited a joint statement that 66 data protection authorities signed. The statement listed principles for organizations that use content-generation systems, which included:
- Giving people clear, useful details about how AI is being used.
- Making it easy for people to ask for harmful content containing personal information to be removed, and dealing with those requests quickly.
- Putting stronger protections in place for children and giving children understandable information appropriate to their age.
For businesses, that means transparency goes beyond putting an AI label on content. The label may explain how someone produced the material. But that still leaves the question of what happens if the material causes harm to a person depicted or discussed in it.
What People Should Know When Using AI
OPC said there are two cases where people especially need to know that they are dealing with AI. The first involves automated decision systems, which use software to produce predictions, recommendations or decisions, or contribute to them, when the results have a significant effect on a person. The second involves a company collecting or using a person’s personal information. OPC said the government could draw on its Guidelines on Obtaining Meaningful Consent as it works out its approach. The guidelines cover the information being collected, who receives it and the reasons for collecting or sharing it, and they also cover the harm or other effects that could result.
OPC said users also need to understand an AI system’s intended purpose and its limits. People using AI themselves need that information, and so do people using AI for business.
AI tools can do powerful things, but those capabilities do not make an AI tool suitable for every job.
For businesses, that is the most important point in the submission. Companies considering AI products need enough information to tell the difference between a tool’s intended job and the other jobs it may appear able to handle. If the explanation is not clear, a buyer could misunderstand the product being purchased, and employees could put an AI system to work on tasks beyond the limits set out for that system.
Questions Remain About AI Failures and Agents
For incidents involving AI, OPC proposed a way for users to report failures and other issues, and said developers should carry out regular tests, which would help make the systems’ known limits clear. OPC included these suggestions in its response to the consultation. They were proposals, rather than an announcement of new requirements.
OPC continues to look at AI agents, which can perform tasks. This includes research and international discussions through the Global Privacy Assembly, and OPC said more information could be shared as that work progresses.
OPC also supported giving people explanations that are accessible and understandable. OPC said future transparency work should make it an explicit aim to give people information that helps them make informed decisions, and to provide it in a format they understand and at a time when they can use it.
The wider issue is whether people can act on what they learn through transparency. Information about an AI system’s problems or limits does less to help someone if it arrives after that person has already relied on the system, and that puts the timing of the information at the centre of the Canadian discussion about trust and AI. The practical question is whether people get the information while there is still time to do something with it.
