The Undefined Word: Ontario's New AI Disclosure and the Recruiter's Risk

Ontario now requires employers to admit when a machine is reading the résumés. Six months in, the rule is live, the definition is a fog, and staffing agencies are holding more of the liability than they may realize.

On January 1, 2026, a short clause buried in Ontario's employment law came into force and quietly changed what every public job posting in the province is supposed to say. The rule sounds almost anodyne: if an employer uses artificial intelligence to screen, assess, or select applicants, the posting must say so. One line of candour, in the name of transparency.

The difficulty is that nearly seven months later, no one can say with confidence where the line falls. Not the employers subject to the rule, not the recruiters writing the postings, and by its own silence not the Ministry that enforces it. For the staffing industry, the ambiguity is not academic. Because of a single phrase in the statute, agencies that post on behalf of their clients are standing squarely inside the obligation.

What the law actually requires

The AI-disclosure rule is one piece of a broader overhaul delivered through Ontario's Working for Workers Four Act and Working for Workers Five Act, now codified in the Employment Standards Act, 2000 at sections 8.1 through 8.6 and fleshed out by O. Reg. 476/24. The package applies to employers with 25 or more employees and governs any "external job posting that an employer or a person acting on behalf of an employer advertises to the general public." Internal postings, and roles performed outside Ontario, fall outside it.

Taken together, the new obligations reshape the anatomy of a compliant posting. Employers must now publish an expected salary or a compensation range, and the range cannot span more than $50,000, which ends the practice of advertising a role at "$60,000 to $120,000" and calling it transparency. Compensation above $200,000 is exempt. Postings must disclose whether they represent a genuine, existing vacancy or are speculative pipeline-building. Employers may no longer require "Canadian experience" in a posting or on an application form. Interviewed candidates must be told the outcome within 45 days, and the postings and applications themselves must be retained for three years.

And then there is the AI clause, the one that has generated the most anxiety and the least clarity.

A definition wide enough to swallow the tool shed

The statute does not define artificial intelligence in the narrow sense a technologist might expect. It reaches for something far broader: "a machine-based system that, for explicit or implicit objectives, infers from the input it receives in order to generate outputs," outputs that include predictions, content, recommendations, or decisions. If such a system is used to "screen, assess, or select" applicants, the posting must disclose it.

Read literally, that language stretches well past the frontier models most people picture. An applicant-tracking system that ranks résumés by keyword match infers from input to generate a recommendation. A scheduling tool that shortlists candidates by availability generates a decision. As the law firm Osler observed, the Ministry "has yet to issue guidance on what is captured by the broad definition of 'artificial intelligence,'" leaving employers and recruiters to construct their own defensible interpretations and hope the interpretation holds.

This is the rare compliance obligation where the safest posture and the most honest posture may diverge. Disclose too little and you risk a breach. Disclose on everything, reflexively, and the notice becomes noise, a boilerplate line appended to every posting that tells applicants nothing and satisfies the spirit of the rule not at all. Neither the over-discloser nor the under-discloser can point to a rulebook that vindicates the choice, because the rulebook, for now, is a single sentence and a regulatory shrug.

Why this lands on the staffing desk

Most coverage of the new rules has framed them as an employer problem. For the staffing industry, that framing misses the phrase that matters most: "or a person acting on behalf of an employer."

That is what an agency is. When a recruiter drafts and publishes a posting for a client, the agency is the person acting on behalf of the employer, and the disclosure obligations travel with the posting. The practical consequence is a liability that sits in an awkward place. The AI in question is often not the agency's at all. It may be the client's proprietary screening model, applied downstream after the agency hands over the shortlist. Yet the disclosure has to appear at the top of the funnel, on the posting the agency wrote, before anyone knows which tools the client will bring to bear.

An agency, in other words, can be required to disclose the use of a system it does not own, cannot inspect, and may not even be told about. The reverse exposure is just as real: an agency running its own AI-assisted ranking inside its applicant-tracking system carries the obligation whether or not the client ever asks. At the volume a busy desk posts, dozens of live roles at once, a single omitted line, replicated across a template, becomes a systematic breach rather than a one-off slip.

For firms already navigating Ontario's temporary-help-agency and recruiter licensing regime, this is another compliance surface layered onto a business model built on posting at scale. The exposure is not evenly distributed across the labour market. It concentrates, by design, on the intermediaries.