If you employ 25 people and post a job, Ontario has a rule you may have missed

Since January 1, 2026, an Ontario employer with 25 or more employees must state in every publicly advertised job posting whether artificial intelligence is used to screen, assess or select applicants. It comes from amendments to the Employment Standards Act, 2000 and Ontario Regulation 476/24, delivered through the Working for Workers legislation.

The disclosure itself is one sentence. What catches employers is not the wording. It is discovering that their applicant tracking system has been doing exactly the thing that triggers the duty.

The rule in plain terms

Three conditions have to line up.

1. The employer size. 25 or more employees, counted on the day the posting goes up.
2. The posting type. Publicly advertised. A vacancy circulated only inside the organisation sits outside this particular duty.
3. The activity. Artificial intelligence used to screen, assess or select applicants.

Where all three apply, the posting must say so. It does not have to explain the tool, name the vendor, describe the model, or set out how candidates are weighted. A compliant disclosure looks close to this:

We use automated tools, including artificial intelligence, to review and screen applications for this role.

That is the whole obligation on the face of it. The difficulty is in the third condition.

The part that actually catches people

The Ministry has not published detailed guidance on how broadly “artificial intelligence” is drawn, or on where “screen”, “assess” and “select” begin and end. Employment lawyers writing about the amendments have flagged exactly that ambiguity.

Meanwhile most employers over 25 people use an applicant tracking system, and a large share of those systems do at least one of the following by default:

• Score or rank candidates against the job description.
• Sort applications by a fit or match percentage.
• Filter out applications that fall below a threshold.
• Parse a resume and infer skills that the candidate did not list.
• Suggest a shortlist.

Every one of those is closer to “screen, assess or select” than to neutral record keeping. And in most businesses nobody chose them. They arrived switched on.

So the practical first step is not writing a sentence. It is finding out what your own system does.

A one-afternoon compliance check

Ask your vendor, in writing: does this product use machine learning, scoring, ranking, matching or automated filtering at any point between an application arriving and a human reading it? Ask for the answer by email. A verbal assurance about a statutory obligation is worth nothing later.

Check the settings yourself. Look for anything called match score, fit, ranking, knockout question, auto-reject or recommended candidates. Note what is on.

Check the tools around the system. Screening questionnaires with automated scoring, video interview platforms with automated assessment, and resume parsers all sit inside the same question even when they are separate products from separate suppliers.

Count your employees on the posting date. For a seasonal Ontario employer this genuinely moves. A business at 22 people in February and 60 in June crosses the threshold partway through its own hiring cycle, and the postings that go up after it crosses are the ones that carry the duty.

Then decide the sentence, and put it in the template. Not in each posting by hand. In the template, so it cannot be forgotten by whoever is covering for the person who usually posts jobs.

Keep a record of the decision, dated. Once you have looked at this, “we did not know” is no longer available. That is not a reason to avoid looking. It is a reason to write down what you found and when.

Why this is the visible edge of a larger obligation

The job posting rule is narrow and specific, which makes it a useful prompt. If AI is touching hiring, it is almost certainly touching other things in the same business, and hiring is the one place where a person is affected by a decision and can ask about it.

The wider exposure is privacy. In May 2026 the Office of the Privacy Commissioner of Canada found that OpenAI’s initial training of ChatGPT did not comply with Canadian privacy law, and complaints under PIPEDA rose 109% year over year to 3,044. Candidate applications are personal information belonging to a person who did not choose your tooling. That reasoning is set out in what PIPEDA means for your AI tools.

The two questions worth asking alongside the posting disclosure:

• Does any candidate information leave your systems, and where does it go?
• Is anybody in the business pasting a resume into a general chatbot to summarise it?

The second happens constantly, it is almost never authorised, and it is a disclosure of somebody else’s personal information by your organisation. That specific problem is covered in your staff are already using AI.

What a written policy should cover

A hiring-specific rule is not enough on its own, because the same staff use the same tools for other work. A workable AI use policy for an Ontario employer covers:

• Which tools are approved and which are specifically not, named rather than described in categories.
• What may never be entered into a general chatbot: candidate and client personal information, anything under a confidentiality agreement, credentials, unreleased commercial terms.
• Which outputs need a human check before they leave the building, and who that human is.
• Disclosure obligations, including the job posting duty and anything a customer contract requires.
• What happens when somebody breaks the rule, written as a process rather than as a threat.
• A dated record of the decision.

That is the scope of AI policy and staff training on this site, at $1,500 to $4,000 including a training half day. It is the cheapest engagement here and the one most Ontario employers over 25 people need first, ahead of any build.

Three things this rule is not

It is not a ban. Using AI in hiring remains lawful. The obligation is to say that you do.

It is not a technical standard. Nothing requires you to audit the tool for bias, publish its logic, or certify it. Those obligations may arrive later. They are not what came into force on January 1, 2026.

It is not the only thing that changed. The same amendments brought other job posting requirements into force on the same date, including obligations around pay information and record keeping. If you are reviewing postings for the AI disclosure, review them once for everything rather than twice.

There is a fourth thing it is not, which is a reason to stop using the tools. Screening software exists because a posting for a general role in Toronto or Mississauga can attract several hundred applications and somebody has to reduce them to a shortlist. Ontario has not said that is improper. It has said the people applying are entitled to know it is happening, which is a much smaller ask than the compliance anxiety around it suggests.

The short version

Count your employees on the day you post. If it is 25 or more and the posting is public, find out what your applicant tracking system does before you decide the answer is no. Then put one sentence in the template, write down the date you checked, and treat it as the prompt to write the AI use policy you were going to need anyway.

This is a summary of published law rather than legal advice, and this area is being actively amended. Verify anything you intend to rely on against the current regulation or with an employment lawyer.