More and more people in Quebec are finding out that their first contact with a debt collector might not be a phone call from a real person at all. Instead, it’s usually a neutral chatbot sending them a message at the right time of day when they’re most likely to respond. That kind of targeting isn’t random; it’s put together by an algorithm.
The Autorité des marchés financiers, Quebec’s financial regulator, is looking into how these AI tools are being used to collect consumer debt in response to this change. Automation in finance isn’t new, but the tools themselves have gotten a lot better. They can now quietly sort through personal data to create responses that sound like a conversation but are completely machine-driven.
Quebec AI Debt Collection Probe – Key Facts (2026)
| Regulatory Body | Autorité des marchés financiers (AMF) |
|---|---|
| Focus of Investigation | Use of AI systems in debt collection across Quebec’s financial sector |
| Guidelines Timeline | Draft published in 2025; public consultation closed Nov 7, 2025 |
| Main Regulatory Concerns | Operational risk, governance, data privacy, algorithmic transparency |
| Compliance Requirements | AI system registries, human oversight, bias mitigation, audit trails |
| Privacy Oversight | Monitored in coordination with Quebec’s privacy commission |
| Technologies Under Review | Predictive analytics, chatbots (e.g. WhatsApp bots), automated outreach |
| Legal Frameworks Referenced | FDCPA, Reg F, TCPA, Quebec’s privacy legislation |
| Key Regulatory Goal | Ensure AI tools treat clients fairly while maintaining legal compliance |
The regulator doesn’t want to stop innovation; they just want to set some rules before the ink dries. The AMF has taken a forward-looking approach by releasing draft guidelines in 2025 and holding a public consultation process that ended in November. It wants institutions to see AI as more than just a convenience; it wants them to see it as a regulated system that needs paperwork, risk analysis, and, most importantly, human responsibility.
These aren’t risks that could happen. A lot of AI-powered debt recovery platforms look at more than just whether someone is behind on a payment. They also look at how that person might respond to different types of outreach. That means that predictive analytics are customizing the timing, tone, channel, and frequency of messages. For example, they can tell if a borrower is more likely to respond to a soft text at 8:42 a.m. on a Tuesday than a firm email on Friday afternoon. This brings up important questions about ethics and free will. When does personalization begin to feel like pressure?
It’s smart for regulators to be careful. The AMF’s proposed framework says that banks and other financial institutions must keep a full list of all the AI systems they use in their work, even the ones that help them talk to customers, like when they collect debts. These systems need to be checked on a regular basis, including tests for bias and how well they work in different situations. It’s a very disciplined way to do things, especially since the rules are still catching up with how quickly AI is being used.
The part of the guidelines that requires human oversight is probably the most important. Every AI system that interacts with customers must be able to be explained, and its decisions must be able to be checked. This means that if an algorithm suggests a payment plan, a person should be able to explain why the machine came to that conclusion instead of just trusting that it “learned” it correctly.
There is also a strong privacy aspect. Quebec’s privacy watchdog has raised concerns about how collection algorithms are getting and using consumer data. A lot of systems now check external datasets to see how likely someone is to pay back a loan. They may even use demographic, behavioral, or location data. If you don’t use these inputs correctly, they can make things much less fair.
A compliance officer from a midsize bank said at a fintech forum in Laval last fall that they weren’t always sure where their vendor’s chatbot data was stored or how often it was retrained. “We get dashboards,” she said, “but we can’t always see what’s behind them.” That’s the kind of lack of transparency that regulators are trying to fix.
Last year, I watched a demo where a collections bot pretended to talk to a fake debtor five times. The language changed a little bit each time. The wording became so exact by the fifth message that it felt like it was made for someone’s exact mental state. It worked very well, but it was also a little creepy.
But not all of the responses to the AMF’s push have been defensive. Some lenders see the change in regulations as a chance to gain more trust from customers. They want to show that automation can still be kind by making sure their AI tools work with these new standards. One credit union in Montreal has already started giving borrowers transparency reports that explain how their debt interactions were decided. This has been especially helpful for clients who were hesitant to use digital services before.
It will also be important for the law to be in line. Quebec’s AI tools must follow national rules like the Fair Debt Collection Practices Act, the Telephone Consumer Protection Act, and Regulation F, which sets rules for how many times a consumer can be contacted in a certain amount of time or whether they have chosen not to receive certain types of communication. Quebec also has strict rules for protecting data, which require meaningful consent and limit automated profiling unless the law allows it.
The AMF’s overall approach encourages institutions to combine technical knowledge with moral responsibility. It’s a good reminder that a system shouldn’t be able to run without limits just because it can figure out the best outcomes. Especially not when people’s jobs and dignity are at stake.
Some vendors are already changing the products they sell. One AI platform that used to focus on “nudge-based repayment strategies” has changed its name to put more emphasis on being open. Another has added built-in compliance alerts that let you know when the frequency of messages is getting close to legal limits. This is a very effective way to stop overreach.
The bigger conversation in Quebec is changing how we think about AI accountability—not by putting in place strict rules, but by setting clear expectations. It could be a model for other provinces or even the federal government to follow.
AI is changing the way businesses interact with customers all the time, and rules like this add an important layer of stability. It strikes a balance between caution and hope: it pushes for progress, but never at the cost of clarity, legality, or fairness.
