A new federal privacy bill is headed for a second reading this month, and it says more about your AI agent, your voice receptionist and your automated workflows than its name suggests.
The short answer: Bill C-36 is not law. It had first reading on 15 June 2026 and sits at second reading, waiting for Parliament's return on 21 September 2026, and there is no firm timeline for royal assent or coming-into-force. It does not create an AI-specific regime — AIDA died with the earlier Bill C-27 — but it rewrites the privacy rules that already apply to anything an AI agent touches: inferred data, automated decisions, sensitive information and cross-border transfers. If you run or are planning an AI receptionist, agent or workflow automation in Canada, the right move now is not to wait for a deadline. It's to build the data map, the human-review path and the Canadian-residency architecture the bill will eventually require, because that architecture is worth having regardless of when the bill passes.
Bill C-36, formally An Act to enact the Protecting Privacy and Consumer Data Act (PPCDA), had first reading in the House of Commons on 15 June 2026, sponsored by Evan Solomon, Minister of Artificial Intelligence and Digital Innovation (Government of Canada backgrounder, 15 June 2026). It repeals Part 1 of PIPEDA and replaces it with the PPCDA, and it creates the Digital Safety and Data Protection Commission of Canada, a Commission that can issue binding orders, with a designated Privacy and Consumer Data Commissioner inside it, rather than the separate tribunal C-27 proposed. The government describes it as "the most significant change to Canada's private-sector privacy law in over 25 years" and frames it as "a key element in advancing Canada's National AI Strategy – AI for All."
As of 8 September 2026, LEGISinfo lists the bill as at second reading in the House of Commons, with no second-reading activity recorded yet (LEGISinfo, Bill C-36). Parliament rose for summer on 18 June 2026 and is scheduled to return 21 September 2026, which is when second-reading debate is expected to resume (Blakes, 24 June 2026). This is the third attempt at this reform: Bill C-11 (2020) and Bill C-27 (2022) both died on the Order Paper before reaching royal assent. Unlike C-27, C-36 does not bundle in a stand-alone AI act — Fasken puts it plainly: "Bill C-36 does not include stand-alone AI regulation and is solely focused on privacy" (Fasken, 18 June 2026).
The headline numbers get the attention — fines up to $25 million or 5% of global revenue — but the provision that should worry a small or mid-size business more is section 132, the private right of action. Once a Commissioner has made a finding, a court has ruled, an appeal is exhausted, or a compliance agreement without a damages component has been reached, any individual affected by a contravention can sue for damages in Federal Court or a provincial superior court, as a full civil action with pleadings, discovery and trial, within two years of becoming aware of the triggering event.
McCarthy Tétrault's TechLex team — Connor Bildfell, Jade Buchanan and Gillian P. Kerr — published an analysis on 2 September 2026 that is the reason this topic is timely right now. Their read: "This new private right of action could increase proposed class action filings," and, more pointedly, "In some ways, this proposed right of action, and the class action risk it carries, may pose an even greater threat than PPCDA's massive potential administrative monetary penalties" (McCarthy Tétrault, 2 September 2026). Their strongest line is worth sitting with if you run any system that processes customer data at scale: "even a single complaint by a single individual about a single practice can potentially become the first step in a putative class action." An AI agent that talks to hundreds of customers a month is exactly the kind of single practice they mean.
It is accurate and important to say Bill C-36 is not an AI law. In practice that distinction barely helps a business running AI agents, because the bill's ordinary privacy machinery now explicitly covers the things AI systems generate and use. Personal information under the PPCDA expressly includes information "inferred" about an individual, so a profile, a risk score or a recommendation your AI agent produces about a customer is personal information in its own right, not just the raw inputs that fed it (Michael Geist, 18 June 2026).
The bill also names a list of sensitive information categories that carry extra safeguard, retention and consent obligations: a child's personal information, racial or ethnic origin, political opinions, religious or philosophical beliefs, trade-union membership, genetic information, health information, biometric information capable of uniquely identifying an individual, and sexual orientation. And it extends the "legitimate interest" exception (with a documented privacy impact assessment) to cover disclosure as well as collection and use — relevant the moment your AI workflow shares data with a booking system, a CRM, or a subcontracted service provider, which the bill permits without fresh consent as long as the service provider safeguards the data and reports breaches back to you.
Section 63 is the provision most directly aimed at AI. It covers technology that assists or replaces human judgment — rules-based systems, regression, predictive analytics, machine learning, deep learning and neural networks all qualify. The trigger is not "did AI touch this," it's the effect of the output: for a prediction, recommendation or decision with a "legal or similarly significant effect" on an individual, an organization must publish a general account of how it uses such systems and, on request, explain the type of personal information used, its source, and the principal factors in the outcome. The individual can also make written representations to a human able to review the decision. This is a narrower, GDPR-style threshold than C-27's earlier "significant impact" language.
For an SMB running AI agents, the practical line is usually clear once you draw it. An AI receptionist that books an appointment, reschedules a visit, or answers a billing question is not making a decision with a legal or similarly significant effect — it's executing a routine transaction the customer asked for. An AI agent that independently declines a loan application, screens out a rental applicant, or rejects a job candidate at the resume stage is a different animal: that is exactly the kind of decision section 63 is built for, and it needs an explanation template and a human reviewer in the loop before it ships, not after a complaint arrives.
Voice deserves its own line because it's easy to get this one wrong in both directions. The bill does not say every call recording is biometric information. It says biometric information "capable of uniquely identifying an individual" is sensitive. A recorded phone call becomes biometric-adjacent when it is used, or analyzed, to identify a specific caller by their voice — not simply because a voice was recorded. Health information is unambiguously sensitive regardless of the medium, so a call where a patient describes symptoms or confirms a diagnosis to book an appointment is sensitive information the moment it's captured, whether by a human receptionist or an AI one. The practical implication for a clinic running an AI voice receptionist is the same either way: treat call content involving health information as sensitive from the first second of the call, log access to it, and don't hold it longer than the retention window you can justify.
Section 57 requires a privacy impact assessment and mitigation — contractual protections, approved codes or certification — before personal information is transferred outside Canada, with the assessment producible to the Commission on request. The government's own framing is direct: the bill is "requiring that organizations assess and mitigate any privacy risks before sending personal information outside of Canada."
This is the provision that makes architecture, not policy language, the real compliance work. Maya, our AI voice receptionist for Ontario clinics and SMBs, is built the way this section expects a customer-facing AI system to be built before it's asked to be: data lives in a Supabase Canada region with row-level security and per-tenant isolation, and the voice orchestration layer, Retell AI, operates under a signed business associate agreement. That doesn't make Maya "certified" under a bill that doesn't yet exist as law — no one can claim that — but it means the cross-border question the bill is asking has already been answered for that deployment rather than left for a future audit to discover.
Ontario has no general private-sector privacy statute of its own; PIPEDA (soon the PPCDA) covers commercial activity in the province the way it does everywhere else PIPEDA already applies. PHIPA, in force since 2004, separately governs health information custodians — dental practices, physiotherapy clinics, physicians — and Bill C-36 does not touch it or replace it. A clinic deploying AI ends up in a federal-plus-provincial stack: PHIPA sets the custodian-specific rules for health information, and the PPCDA sets the commercial-activity baseline underneath it, with health information additionally flagged as a sensitive category requiring extra safeguards. Quebec (Law 25), British Columbia and Alberta (each with a PIPA) have their own substantially similar private-sector laws that stand in for the federal statute there. The Ontario IPC has already published guidance on AI in the health sector worth reading alongside this, and our dental industry page and healthcare AI receptionist guide go further into what that combination looks like operationally.
None of the following requires the bill to pass. All of it is good practice today and becomes the compliance baseline the moment it does.
The proposed penalties are real numbers in the bill text: administrative monetary penalties up to $10 million or 3% of global revenue, whichever is greater, and fines up to $25 million or 5% of global revenue for the most serious offences, both straight from the government's own backgrounder. What is not real yet is a compliance deadline. Coming-into-force is tied to standing up the new Commission, which itself depends on the separate Bill C-34. Michael Geist's assessment, offered as an estimate rather than a fact, is blunt: "rather than updating the law for 2027, it is updating it for 2030 or later." Treat any "Bill C-36 compliance deadline" claim with skepticism — there isn't one yet, and there may not be for years. Build the architecture now anyway: a privacy management program, a data map and Canadian data residency are the same build whether royal assent lands in 2027 or 2030, and retrofitting them after a complaint or a class action costs more than building them in from the start.
This article is general information, not legal advice — talk to a privacy lawyer about your specific obligations.
We're not a law firm and we don't claim any certification under a bill that hasn't passed. What we do is build AI agents and AI workflow automation the way this bill expects customer-facing AI to be built: Canadian data residency, documented vendor relationships, and human-review paths for anything that could plausibly be a significant decision. If you're evaluating where your current AI stack stands, the practical first step is our AI Readiness Sprint (CA$5,000–9,500) — a focused engagement that maps your data flows, flags automated decisions that need an explanation template, and hands you a plan before you spend on a build. For teams ready to deploy, our AI agent development work moves from pilot to production (CA$30,000–60,000) with the architecture above built in from day one, and for businesses that want an ongoing owner of this without a full-time hire, a Fractional AI Lead (CA$3,500–7,500/month) keeps the privacy posture current as both your AI stack and the bill itself evolve. Read more about our overall approach on responsible AI in Canada, and see how we evaluate outside vendors in our guide to vetting an AI automation agency or what a build typically costs in our AI agent cost guide.
Build the data map, the explanation templates and the Canadian residency architecture now, before a deadline forces it.