Google will begin using artificial intelligence to estimate the ages of Canadian account holders within days, the company said Monday, inferring whether a user is under 18 from the things they have searched for and the categories of videos they have watched on YouTube.
“Our age estimation model uses machine learning to interpret a variety of signals already associated with a user’s account,” the company said in a blog post, according to The Canadian Press. Jeanette Patell, director of government affairs and public policy at Google Canada, wrote that the rollout would be gradual. If the system flags an account as belonging to a minor, Google says it will apply additional protections, including safeguards on content recommendations. If it wrongly flags an adult, that person can prove their age using a government ID or a selfie. The company says it will collect no new information beyond what is already tied to the account.
What the wire coverage did not mention is that Canada already has a published set of expectations for exactly this technology, and that several of them point in a different direction than the model Google is deploying.
On May 4, 2026, Privacy Commissioner Philippe Dufresne released two guidance documents on age assurance, one for websites and online services and one for the developers who build these systems. The guidance is not a statute. It interprets the reasonable-purpose test in section 5(3) of the Personal Information Protection and Electronic Documents Act, the federal private-sector privacy law that governs Google in Canada today, with or without new legislation. It adopts the international standard definition of age assurance, which expressly covers processes that verify, estimate or infer a person’s age. An inference model trained on search and viewing history is squarely inside that scope.
The guidance sets out a principle, borrowed from the Spanish data protection authority, that age assurance should be aimed at having users prove they are authorized to access something, and not at verifying who is a minor. Google’s system is built the other way around. It runs across the existing account base looking for children, then applies restrictions to the accounts it identifies. The guidance also states plainly that age assurance should not be the default condition for accessing the internet, and that organizations should limit how often a person is put through the process.
There is a second tension in the appeal path. The OPC guidance requires that anyone denied access on the basis of an age result have an appeal mechanism, and that the mechanism itself be privacy-protective and proportionate to the risk. Google’s remedy for a false positive is a government ID or a selfie, which is the most data-intensive option available. Under that design, the cost of the system’s own error falls on the user, in the form of the identity document the inference model was supposed to make unnecessary.
The third question is the one Google’s statement does not address at all. The guidance says an age assurance result must not be used for any other purpose, must not be used to correlate a user’s visits, and should be destroyed as soon as possible. It also identifies detailed behavioural profiling as a practice that can itself be harmful to children. Google’s age signal is not generated beside the advertising profile. It is generated out of it, from the same search and watch history. The company did not reference the Canadian guidance in the remarks quoted by The Canadian Press.
All of this is landing before Parliament has decided anything. The Liberal government introduced Bill C-34, the Safe Social Media Act, in June. It would require platforms to at least temporarily block users under 16, but it is not law, and the government has said it has not settled how age verification would work and will consult platforms on the methods. Manitoba has separately said it intends to become the first province to restrict social media for young people.
That sequencing matters. By the time MPs vote, the largest search and video company in the country will already have shipped a working answer to the question the bill is meant to settle, and the terms it sets now will be hard to unset later. Patell’s post made the company’s preference explicit, arguing against “a universal arbiter of age” and for liability resting with individual developers and publishers rather than at a single chokepoint. That is a position with commercial consequences, and it is being advanced while the consultation is still open.
One more gap is worth noting. Meta already uses AI to place suspected minors into teen accounts in Canada, but the child-protection measures it agreed to as part of a major United States legal settlement are not being extended here, the company has confirmed. Canadian users are getting the detection, not the settlement.
Reported via The Canadian Press (Anja Karadeglija, Sept. 14, 2026). Additional reporting from the Office of the Privacy Commissioner of Canada’s age assurance guidance, published May 4, 2026.
