Ontario AI Hiring Disclosure in 2026: What Counts?

Since January 1, 2026, Ontario employers with at least 25 employees generally must state in a public job posting when artificial intelligence is used to screen, assess, or select applicants. The employer need not name or explain the system, and remains responsible when a recruiting firm uses AI on its behalf.

The disclosure sentence is the easy part. The harder job is discovering what your configured ATS features, assessments, interview tools, and recruiting partners actually do. A product name cannot answer that question. You need to trace how each feature handles applicant data and whether its output affects who progresses.

This guide explains the statutory minimum and offers a practical audit. It is general information, not legal advice. The Ontario rule is new, several important terms remain undefined, and fact-specific questions should go to employment counsel.

What must Ontario employers disclose about AI in 2026?

Section 8.4 of Ontario’s Employment Standards Act, 2000 requires a covered employer to include a statement in a publicly advertised job posting if it uses AI to screen, assess, or select applicants for the position. The requirement is part of the ESA, not a standalone “Ontario AI Act.”

The statement belongs in the public job posting. It does not belong only in the application form, privacy notice, or recruiting platform terms. Ontario’s Ministry guidance says you do not need to provide a detailed description of the system or explain how it is used. A direct sentence stating the relevant use is enough.

The trigger is the use of AI for one of three functions:

• screening applicants;
• assessing applicants; or
• selecting applicants.

The law does not expressly require a statement that you do not use AI. You may publish one voluntarily, but it is not the statutory requirement in section 8.4.

Which employers and job postings are covered?

The regulation generally exempts an employer that employs fewer than 25 employees on the day the posting is posted. That means exactly 25 is within scope. The Ministry guide describes the count as 25 or more Ontario employees and counts individuals, not full-time equivalents. A part-time or casual employee counts as one person.

Do not substitute global company headcount for the Ontario test. Federal-jurisdiction employment relationships, certain Crown bodies, and other scope questions can complicate the answer. If your organization sits near a boundary, confirm the count and jurisdiction with counsel before publishing.

The regulation defines a publicly advertised job posting as an external advertisement to the general public for a specific position, whether the employer or its agent publishes it. The definition excludes:

• a general recruitment campaign that does not advertise a specific position;
• a general help-wanted sign that does not advertise a specific position;
• a posting restricted to existing employees; and
• specified positions whose work is outside Ontario.

An external posting on your careers site, a job board, or a recruiting agency’s site can therefore qualify. Using an agent does not transfer the obligation. If a recruiting firm publishes the posting or uses AI on your behalf, you still need enough information to decide what the posting must say.

What counts as AI screening, assessment, or selection?

Ontario defines AI as a machine-based system that infers from its inputs to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Inference and effect matter more than the product label. An ATS is not automatically covered, and an innocently named “assistant” is not automatically outside the rule.

The ESA and regulation do not define “screen,” “assess,” or “select.” No published decision applying section 8.4 was identified in the research for this article. The matrix below is therefore an operational interpretation, not settled law.

Use a three-part test for every configured feature:

1. Is it inference-based AI under the regulation? Identify what inputs it receives and what output it generates.
2. Does it act on applicants for this position? Separate applicant evaluation from sourcing, content creation, and administration.
3. Does the output filter, rank, score, evaluate, recommend, or influence progression? Look at actual workflow effects, not vendor marketing.

When all three answers are yes, disclosure is the safer reading. A recruiter making the final decision does not undo the earlier use of AI. The enacted rule contains no human-review exemption, although a court has not yet settled every boundary.

Does your ATS, resume parser, or interview tool trigger disclosure?

You cannot classify a tool once and reuse that answer across every role. The same platform can be administrative in one workflow and evaluative in another. Configuration, data flow, and recruiter behaviour determine the useful compliance answer.

Consider a resume parser. If it only copies a candidate’s stated employer and job title into structured fields, the result may be extraction rather than inference-based evaluation. If it infers skills, scores experience, recommends matches, or feeds an automated shortlist, the facts point toward screening or assessment.

A fixed rule also needs examination. A simple checkbox that removes anyone who states they lack a legally required licence may be deterministic rather than AI. A system that predicts whether a resume demonstrates the licence, weighs related experience, and ranks the pool is a different process.

Recorded-interview software is another clear example. Storage and playback are administrative. Transcription alone may not evaluate anyone. Scoring answers, inferring traits, creating a suitability score, or recommending who advances is assessment or selection in ordinary operational terms.

Human review does not provide a general escape hatch. If an AI ranking changes which applications recruiters open first, shapes the evidence they see, or recommends who advances, the system may influence screening or assessment even though a person clicks the final button.

What should an Ontario AI disclosure statement say?

Your statement should be accurate, direct, and matched to the configured workflow. The statutory minimum can be short. Candidate-friendly wording may add useful context, but every extra promise must remain true in practice.

Minimal screening statement

We use artificial intelligence to screen applicants for this position.

Candidate-friendly screening statement

We use AI to compare applications with the stated job requirements during initial screening. Recruiters review the results and decide who progresses.

Assessment statement

We use AI to assist in scoring written assessment responses. Hiring staff review the assessment and make all progression and hiring decisions.

Third-party statement

Our recruiting partner uses AI to rank applications against the job criteria. Our hiring team reviews the ranked applications and makes all decisions.

Do not write “AI may be used” if AI is in fact used in the workflow. Do not promise recruiter review, a particular decision process, or a limited use unless operations match the sentence. If different roles use different tools, publish role-specific wording rather than one vague company-wide disclaimer.

Keep the sentence consistent across every public channel. If your careers site has the disclosure but the recruiting firm’s copy does not, candidates are still seeing a public version without it. Treat the posting as a versioned record distributed to several destinations.

What does Ontario’s AI hiring rule not require?

Section 8.4 is a disclosure rule, not a complete responsible-AI framework. It does not itself require you to provide:

• an AI ban, applicant consent, opt-out, or applicant-specific notice;
• the vendor name, model, logic, criteria, data, score, or automation level;
• a bias audit, accuracy test, impact assessment, public report, or appeal;
• a mandated human-in-the-loop process;
• a negative statement when the disclosure trigger is absent;
• disclosure for AI used only in unrelated administration or content drafting; or
• three-year retention of applications, resumes, AI scores, or an AI inventory under these specific posting-record provisions.

Those boundaries do not make a disclosed practice lawful or responsible by default. Human rights, privacy, accessibility, and other employment rules may apply independently. The Ontario Human Rights Commission and Information and Privacy Commissioner have published responsible-AI principles, but those principles are governance guidance, not additional text in the ESA posting requirement for private employers.

The distinction matters. Label your controls accurately: the disclosure is a legal duty when triggered; an impact review, performance test, or appeal process may be prudent governance or required under another rule. Do not present the legal minimum as a safe harbour.

How do you audit a hiring workflow for AI use?

A useful audit follows the applicant’s data, not your software list. The public statement and prescribed retention are legal duties under the cited provisions. The inventory, vendor evidence, and review controls below are prudent ways to reach and support the right decision.

1. Inventory by function

List every ATS feature, parser, assessment, chatbot, interview tool, model, and recruiting partner used for the role. Break suites into enabled features. “We use Vendor X” is not specific enough when only one module ranks applicants.

2. Map the actual data path

For each feature, record its input, output, inference, and workflow effect. Ask whether the output filters, ranks, scores, evaluates, recommends, changes progression, or merely stores and transports information. Include manual workarounds, integrations, and exports that may not appear in the main ATS diagram.

3. Get the vendor answer in writing

Ask which features are enabled, what system produces each output, whether subcontractors use AI, and whether settings can change without notice. Ask about behaviour, not labels. A vendor saying “we are AI-powered” or “we do not make decisions” does not reveal whether its ranking influences your recruiters.

4. Publish accurate wording and preserve the version

Put the statement in the public posting and repeat it on each channel. Preserve the complete public version, its linked content, and its revisions. A dated tool inventory, vendor response, configuration export or screenshot, owner approval, posting snapshot, version history, and change-review date can help explain how you reached the wording.

5. Review every workflow or vendor change

Recheck the decision when someone enables a feature, changes an integration, adds a recruiting partner, or sends the role through a new assessment path. Assign an owner and review date. Without change control, a correct disclosure can quietly become inaccurate.

Industry language is already changing quickly. Indeed Hiring Lab reported that AI-related terms appeared in 28% of the Ontario postings in its tracked sample in May 2026, up from 9% in October 2025. That is not a population estimate of actual AI use, because some terms may describe job skills, but it shows why a one-time software audit will age badly.

Which other Ontario job-posting duties apply?

AI disclosure is one part of Ontario’s 2026 public-posting regime. Covered postings also generally require expected compensation or a range, subject to the over-$200,000 exception and a maximum advertised range width of $50,000; no Canadian-experience requirement; and a statement saying whether an existing vacancy exists.

An employer must also tell an interviewed applicant within 45 days whether a hiring decision has been made. That is not a requirement to provide a rejection, an outcome, or a reason within 45 days. Posting records, the associated blank application form, linked material that forms part of the posting, and public revisions generally must be retained for three years after public access ends.

Ontario did not ban “ghost jobs” or require every advertised vacancy to be filled. For the full set of boundaries and examples, read our companion guide to Ontario’s ghost-job and job-posting rules.

How does Kit separate AI assistance from hiring decisions?

Kit illustrates why a functional inventory is more useful than a broad “uses AI” label. Different features have different boundaries, and your use of their outputs still matters.

Kit’s Live Review is a private assistant for a human recruiter reviewing one candidate. Its safety boundary allows it to surface evidence, unknowns, and questions. It must not advance, reject, rank, score, contact, or claim an employment decision. That boundary is valuable evidence about what the feature cannot do, but it does not decide the Ontario analysis for you.

Kit can also use AI to extract configured metafield values from resumes and application responses. Whether that output becomes screening or assessment depends on how you configure and use it. If an extracted or inferred value filters the pool, prioritizes review, or influences progression, assess the disclosure trigger for that workflow.

Kit’s public application form can display a customizable account-wide notice that a resume may be parsed with AI-assisted tools and preserve the submitted notice with the candidate’s consent. That notice is not a dedicated, role-specific Ontario AI statement in the public posting. It should not be treated as satisfying section 8.4.

Kit retains human-authored reviews and scorecards, but it does not automatically decide whether an Ontario posting is in scope, insert an Ontario disclosure, provide a complete jurisdiction engine, guarantee legal compliance, or promise an immutable three-year archive of every public posting version and linked source. A human-final-decision design also does not remove the disclosure trigger when AI is used earlier to screen or assess.

The practical approach is simple: document each feature’s real function, trace how its output affects applicants, publish a truthful role-specific statement when the trigger is met, and preserve the evidence behind that choice. If you want a more inspectable hiring workflow, start a free Kit trial, then review your Ontario scope and disclosure wording with qualified counsel.