Law 25 and AI Email: How to Choose a Compliant Tool in Quebec (2026)

· Alexandre Sauvageau

Law 25 and AI Email: How to Choose a Compliant Tool in Quebec (2026)

Loi 25 & AI email: 5 red flags in non-compliant vendors, 7 criteria for compliant tools, and 6 questions to ask before you sign. Quebec Law 25 guide.

Using an AI email assistant in Quebec? Under Loi 25, your vendor processes personal data on your behalf — making you liable for their compliance. Here are the 5 red flags to spot and the 7 criteria every compliant AI email tool must meet.

Why your email inbox is personal data under Loi 25

Under Article 2 of Quebec's *Loi sur la protection des renseignements personnels dans le secteur privé* (P-39.1), personal information is any information that, alone or combined with other information, allows a natural person to be identified. Your email inbox qualifies entirely: sender names, email addresses, message content, contact patterns, professional relationships, and behavioral signals (response time, tone, topics discussed) are all personal data. When you connect an AI email tool to your inbox, you are delegating the processing of personal data to a third party. Under Loi 25, you remain the controller — responsible for how that data is handled, stored, and protected. The vendor becomes a processor, bound by your obligations.

This legal reality means that choosing a non-compliant AI email vendor doesn't just create risk for the vendor — it creates direct legal exposure for your organization. If the vendor suffers a breach, you notify the CAI with diligence. If the vendor uses your email data to train their AI models without explicit consent, you have violated Loi 25's purpose limitation principle. If the vendor's data centers are in the United States, you may be facilitating unauthorized cross-border personal data transfers. The controller-processor relationship is not just a legal formality under Quebec law — it is an active compliance obligation that requires vendor due diligence before deployment, not after.

5 Loi 25 red flags to spot in any AI email vendor

Red flag 1: US-only data residency without a qualifying PIA. If the vendor's terms say data is stored in US-based data centers and they cannot produce a Privacy Impact Assessment confirming Loi 25-equivalent protections, that's a compliance gap you own as the controller. The CLOUD Act risk is real: American authorities can compel US companies to produce data stored anywhere globally. Red flag 2: 'We may use your data to improve our models.' This common clause in AI tool terms of service directly violates Loi 25's purpose limitation principle. Your email content was collected for communication — not AI training. Any use beyond the original purpose requires fresh, explicit consent. If the vendor's default terms include model training, that consent was never obtained. Red flag 3: No named Privacy Officer. If you cannot find the vendor's designated Privacy Officer on their website or in their DPA, they are likely not Loi 25 compliant — which makes them a non-qualifying processor under your obligations.

Red flag 4: No Data Processing Agreement (DPA). Loi 25 requires written contracts with every third party that processes personal data on your behalf, specifying privacy protections equivalent to Quebec law. If a vendor doesn't offer a DPA or redirects you to a generic privacy policy, they do not meet the third-party vendor standard under Loi 25. Red flag 5: No disclosure of automated decision-making. Under Phase 3 of Loi 25 (in force since September 2024), any tool using AI to make decisions affecting individuals must disclose that automated processing is occurring and offer the individual the right to human review. If your AI email tool sorts, prioritizes, or drafts without any user-facing disclosure, both you and the vendor may be in violation. Ask vendors specifically: does the tool disclose to email recipients or users that AI processing is involved?

7 criteria a Loi 25-compliant AI email tool must meet

When evaluating any AI email tool for use in a Quebec business context, verify all seven of these criteria before deployment: 1. Canadian data residency. Email content and processed data must be stored in Canadian-certified data centers. This eliminates CLOUD Act exposure and satisfies Loi 25's data residency requirements for sensitive personal information. 2. No model training on user data. The vendor's terms must explicitly state that email content is never used to train AI models, fine-tune foundation models, or improve the service for other users. This is non-negotiable under Loi 25's purpose limitation and consent requirements. 3. A signed Data Processing Agreement. The DPA must specify the nature of processing, purpose, duration, data types, security measures, breach notification timelines, and your right to audit or terminate processing. Available *before* you sign up, not only on request after. 4. Named Privacy Officer published on their website. Verifiable, not just 'privacy@company.com'. A real name with a role. 5. Rapid breach notification commitment. The DPA must specify that the vendor will notify you within 24 hours of discovering a breach so you can meet your duty to notify the CAI with diligence. 6. Human review before sending. No AI email tool should send anything without your explicit approval. Under Loi 25's automated decision-making provisions, users must retain control over consequential actions. 7. Data deletion on request. The vendor must be able to delete all your personal data — emails, voice profiles, behavioral data — within 30 days of a deletion request, with written confirmation.

Agentys meets all seven criteria. Canadian data centers (no US infrastructure). Email content never used for AI training. DPA available before signup. Named Privacy Officer published on the Trust Center. 24-hour breach notification commitment. Nothing sends without your explicit one-click approval. And data deletion executed within 30 days with written confirmation. These aren't compliance checkboxes — they're design decisions made before we wrote the first line of code, because we believe privacy-by-design is the only defensible architecture for a company handling professional email.

6 questions to ask any AI email vendor before you sign

Use this checklist in any vendor evaluation: 1. 'Where exactly is my email data stored?' Require a country-level and data center-level answer. 'North America' is not acceptable. 2. 'Is my email content used to train your AI models?' Require a written answer in the DPA, not just a verbal assurance. 3. 'Can I see your Data Processing Agreement before signing up?' If the DPA is only available post-signup, the vendor is not positioning itself as a Loi 25-compliant processor. 4. 'Who is your Privacy Officer and where can I find them?' If they can't name someone and point you to a public page, they may not have one. 5. 'What is your breach notification SLA to customers?' You need to hear '24 hours' or less. 'As soon as reasonably possible' is a compliance gap. 6. 'How do you disclose automated decision-making to users?' They should describe a specific in-product disclosure, not just reference their privacy policy.

These questions serve a dual purpose: they identify non-compliant vendors and they protect you legally. If a vendor passes all six — in writing, in the DPA — you have documented your due diligence as controller. If a breach or complaint ever arises, your vendor audit records demonstrate that you took your Loi 25 obligations seriously and selected a processor that met the standard of care required by Quebec law. Vendor due diligence is not bureaucracy. It's your liability shield.

Loi 25 doesn't prohibit using AI email tools. It sets a clear standard for how they must be built and how vendors must behave. The seven criteria and six questions above are your complete vendor evaluation framework. Any AI email tool that meets all seven — Canadian data residency, no model training, signed DPA, named Privacy Officer, 24h breach SLA, human review before sending, and data deletion on request — can be deployed in a Loi 25-compliant workflow. Agentys was built to that standard. If you're currently using a tool that can't answer those six questions in writing, you're carrying legal exposure that a single client complaint could activate.