Ontario’s Bill 194 became the Enhancing Digital Security and Trust Act, 2024. It sets four AI duties for public bodies: disclosure, an accountability framework, risk management and human oversight. Those duties apply only once a regulation prescribes them, and none has been announced. Today, the binding AI rule is the Responsible Use of AI Directive for ministries and agencies.
If you run AI in an Ontario hospital, school board, university or municipality, or you sell AI to one, the gap between the Act and the regulations is where your risk sits. Read the Act as already binding and you overbuild. Wait for the regulation and you will build governance under a deadline, after your customers have already put AI clauses in your contract.
What did Bill 194 actually do for AI?
Bill 194 created the Enhancing Digital Security and Trust Act, 2024 (EDSTA). Schedule 1 defines an AI system as a machine-based system that infers from its input to generate outputs such as predictions, content, recommendations or decisions, and McCarthy Tétrault’s analysis of the bill notes the definition reaches third-party tools embedded in government software, not only systems the government builds.
The Act lets Ontario impose four duties on a public sector entity that uses AI in “prescribed circumstances”:
- Tell the public the entity uses AI.
- Develop and implement an accountability framework.
- Manage the risks of harm from the AI system.
- Keep human oversight in the loop.
The Act covers institutions under FIPPA and MFIPPA, children’s aid societies and school boards. The key word is “prescribed.” A regulation must say which AI uses trigger the duties, and it can also prohibit certain uses. The Act sets the frame; the regulation fills it in.
Which AI rules bind Ontario public bodies today?
Only one instrument binds with AI-specific content today, and it does not cover most of the entities the Act names. The Ontario government’s own EDSTA page announces no AI regulation, and Hicks Morley’s March 2026 bulletin states that no AI regulation has been issued under the Act.
| Instrument | Applies to | Binding today? | What it asks for |
|---|---|---|---|
| EDSTA AI provisions | FIPPA/MFIPPA institutions, school boards, children’s aid societies | No — waits for a regulation | Disclosure, accountability framework, risk management, human oversight |
| Responsible Use of AI Directive | Ministries and provincial agencies | Yes, since December 1, 2024 | Risk process, annual use-case reporting, public list, oversight |
| IPC/OHRC Principles (January 2026) | Public-sector institutions | No, but regulators will use them | Six principles across the AI lifecycle |
| O. Reg. 51/26 (cybersecurity) | Hospitals, colleges, universities, school boards, children’s aid societies | Yes, from July 1, 2026 | Security officer, maturity assessments, 72-hour incident reporting |
A school board or hospital therefore has no AI-specific statutory duty today, but it does have a cybersecurity regulation, a privacy regulator watching, and a provincial directive that shows what the coming regulation will probably look like.
What does the Responsible Use of AI Directive require?
The Responsible Use of Artificial Intelligence Directive took effect on December 1, 2024 and applies to Ontario ministries and provincial agencies that use AI to develop or deliver policies, programs or services. It is the closest thing to a working template for the EDSTA duties.
Ministries must run a structured risk process: document the problem, decide whether AI is justified, assess risks, apply proportionate controls, and monitor. They report AI use cases each year to the Ministry of Public and Business Service Delivery and Procurement and publish a list of use cases. They disclose AI when it directly affects a public service or decision, keep “clear ongoing human oversight,” and give the public an accessible way to ask about AI use. Exemptions need Management Board of Cabinet approval through a business case, and the Directive is reviewed at least every two years.
Those requirements map closely to three of the four EDSTA duties: disclosure, risk management and human oversight all appear. If you build to the Directive and add a named accountable owner, you will cover most of what a prescribed-circumstances regulation is likely to ask, though the actual regulation may differ.
What did the July 1, 2026 regulations change?
Two regulations under the EDSTA came into force on July 1, 2026, and neither is about AI. O. Reg. 51/26 requires covered entities to designate a senior cybersecurity contact, run maturity assessments (every two years, per the Ontario government page) and report critical incidents within 72 hours. O. Reg. 52/26 requires school boards to notify parents and students when personal information goes to third-party software providers.
AI matters here for a practical reason. An unsanctioned AI tool that receives staff or student data sits inside the same incident perimeter as any other system. A school board that cannot list its AI tools cannot tell whether a leak through one of them is reportable inside 72 hours. Shadow AI, in other words, becomes a cybersecurity compliance problem before it becomes an AI-regulation problem. Confirm the scope of “critical incident” with counsel, because the regulation’s definition controls.
What do the IPC and OHRC principles add?
In January 2026, Ontario’s Information and Privacy Commissioner and Human Rights Commission jointly issued six principles for responsible AI use: valid and reliable, safe, privacy protective, human rights affirming, transparent, and accountable. They cover the whole lifecycle from design to retirement and reach generative AI, decision systems, and simpler tools such as spam filters.
The principles are recommendations, not law. Lerners reports that both regulators said the principles will “ground our assessment of organizations’ adoption of AI systems.” That turns a soft document into an audit lens. When the IPC or OHRC looks at your AI use, these six headings are the questions. An AI audit trail that shows who approved a system, what data it touched, and who reviewed its outputs answers most of them.
What this means for US and Canadian organizations
Ontario public bodies face a provincial framework with no operative AI regulation yet. The federal picture is thinner: Blakes’ winter 2026 review says broad federal AI legislation is unlikely to return soon, after the proposed AIDA lapsed with Bill C-27. The Treasury Board’s Directive on Automated Decision-Making covers federal institutions, not Ontario entities. Quebec’s Law 25 regulates automated decisions in a separate track, so a national organization needs a province-by-province view.
US vendors selling into Ontario hospitals, universities and school boards feel this through contracts. Denton’s analysis expects public sector clients to pass new requirements to vendors through contract terms, and the 72-hour incident window already reaches anyone who hosts their data.
Both NIST and ISO give you a structure to build against now. The NIST AI RMF functions Govern, Map, Measure and Manage line up with the accountability framework and risk management duties, and an ISO/IEC 42001 management system gives you documented roles, risk assessment and monitoring. Neither substitutes for legal compliance; both give you evidence you can show a customer. See how the AI regulations fit together across frameworks.
What should you do this quarter?
- Build the inventory. List every AI use case with an owner, data category, and vendor. The Directive’s annual reporting requirement is a ready-made field list.
- Sort by impact on people. Flag uses that affect decisions about individuals. The regulation will most likely target those first, but its actual “prescribed circumstances” may differ.
- Draft a public notice template. Write the plain-language disclosure once, so you can publish it within days of a regulation.
- Name an accountable owner and a human review step for each high-impact use.
- Update vendor contracts. Require AI disclosure, logging, and incident notice that fits inside your 72-hour window.
- Watch the Ontario Regulatory Registry for a proposed AI regulation, since proposals are published for comment before they take effect.
A strong AI governance program covers all six steps and answers to whichever regulation arrives.
Key takeaways
- Bill 194’s AI duties (disclosure, accountability framework, risk management, human oversight) are a frame; they bind only when a regulation prescribes the circumstances, and none has been announced.
- The Responsible Use of AI Directive, in force since December 1, 2024, is the only binding AI-specific rule, and it covers ministries and agencies, not hospitals, school boards or municipalities.
- Build to the Directive and the IPC/OHRC principles now: an AI inventory, a risk process, a public notice and named oversight give you the evidence any future regulation will ask for.
Next step: Explore how Fruggr produces audit-ready AI evidence in its AI compliance reporting.
FAQ
Bill 194 became the Enhancing Digital Security and Trust Act, 2024. Its cybersecurity regulations took effect on July 1, 2026. Its AI provisions create duties that apply only once a regulation prescribes the circumstances, and no AI regulation has been announced (verify current status on the Ontario Regulatory Registry).
No. The Act covers public sector entities such as FIPPA and MFIPPA institutions, school boards and children’s aid societies. Private vendors feel it through contract requirements from those customers.
It is a use of AI that a future regulation names as triggering the four duties. Until Ontario publishes that regulation, the term has no operative definition.
No. The Directive applies to Ontario ministries and provincial agencies. Hospitals and school boards fall under the EDSTA’s broader coverage, which has no AI regulation yet.
No, they are recommendations. Both regulators said they will use them to ground their assessment of organizations’ AI adoption, so treat them as an expectation.
This article was drafted with AI assistance and reviewed by our editorial team before publishing.