Responsible AI

Not a pledge.
A practice.

Canada Muse is built in Québec, French-first, under one of North America's strictest privacy laws. Canada has no federal AI statute in force today — the proposed one died on the order paper in January 2025 — so we hold ourselves to the voluntary federal code anyway, and to Québec's Law 25, which is very much in force. Every claim on this page maps to a mechanism you can check.

Built in QuébecLaw 25 appliesThis page is dated
Commitments

What we've put
our name to.

Signing in progress

Montréal Declaration for Responsible AI

Ten principles for responsible AI, born at Université de Montréal in 2018. It is Québec's declaration — and the natural one for a Québec company. Below, all ten are mapped to what we actually run.

Signing in progress

Canada's Voluntary Code of Conduct on Advanced Generative AI

The federal voluntary code (2023): accountability, safety, fairness and equity, transparency, human oversight, validity and robustness. We sign as a manager of systems built on frontier models.

In progress

Law 25 groundwork

Privacy-officer designation, confidentiality-incident register, privacy impact assessment. Underway now — and this page is dated, so each lands here when it is real, not before.

The ten principles

Each principle,
a mechanism.

The Montréal Declaration's ten principles, each paired with something you can verify in the product. Where the honest answer is modest, we give the modest answer.

01Well-beingThe product exists to hand you back your evenings, not to hold your attention. No feed, no streaks, no engagement metrics — and this website loads no tracker unless you say yes.
02AutonomyYou stay the decision-maker. New recipients, outward commitments and spending gate on your explicit yes. Credits are prepaid; nothing is charged automatically unless you switch auto-recharge on yourself.
03PrivacyYour workspace is its own database, with row-level security and per-tenant credentials. Voice audio is transcribed and discarded, never stored. Your content never trains a shared model — that's architecture, not a checkbox.
04SolidarityYour client relationships stay human. The assistant writes in your name under rules you set — anyone new hears from it because you said yes, not because a machine decided.
05Democratic participationYou can see how it works. Model routing is disclosed — Anthropic by default, OpenAI and xAI when routed or chosen — actions are logged and explainable, and this page is dated and versioned.
06EquityOne price, stated up front, prepaid, cancel anytime. French is not the discount tier: the product speaks and writes both languages natively, at the same price.
07Diversity & inclusionFrontier models from several providers rather than a single monoculture; a cast of faces and voices you choose from; FR and EN as equals.
08PrudenceUnknown email senders are quarantined until you approve them. Mail that fails authentication is filtered — automatically, and we say so below. Paid provisioning asks before spending a dollar.
09ResponsibilityNo fully automated decision produces a legal or significant effect on a person — a human decides. The two automated behaviours that do exist are disclosed in full below. A privacy officer answers privacy@museai.ca.
10Sustainable developmentWe won't greenwash. We're a small team, and our footprint is mostly the compute behind your requests. We run in AWS's Canadian region and store only what your memory needs — voice audio, for one, is discarded rather than kept.
The two automated behaviours, in full

(1) Inbound email that fails authentication (DMARC) or matches spam rules can be dropped before your assistant sees it. (2) Credit enforcement — currently switched off — can pause paid features when your balance reaches zero, until you top up. That's the whole list; if it grows, this page grows first.

The anti-claims

What we don't do.

The easiest place to be responsible is in the claims you refuse to make.

We don't train shared models on your content.

Not as a policy — as architecture. Your workspace is its own database. There is nothing to opt out of, because there is nothing to opt into.

We don't track you without asking.

No analytics, no pixels and no advertising cookies until you accept the banner. Accept, and Google measures two things: whether an ad led to a sign-up (Google Ads) and how the site is visited (Google Analytics) — ad personalization off, no remarketing. Decline, and nothing loads; change your mind any time from the footer.

We don't sell personal information.

There is no advertising business here to feed. Ad measurement reaches Google only with your yes, and your workspace content never does.

We don't auto-charge.

Credits are prepaid. Auto-recharge exists — and it is off until you, personally, turn it on.

We don't let the assistant commit you to strangers.

For people you already work with, it can act — that's the product. Anyone new hears from it only after your explicit yes.

We don't say “everything stays in Canada.”

Your data is stored in Canada (AWS ca-central-1). To answer you, your messages are processed by US model providers. Saying otherwise would be tidier — and false.

Your rights

Law 25, in
plain words.

Write to privacy@museai.ca — it reaches our privacy officer — to access what we hold about you, correct it, delete it, receive a copy in a structured format, withdraw consent, or complain. We answer within 30 days.

Not satisfied with our answer? The Commission d'accès à l'information (Québec) hears complaints, and so does the Office of the Privacy Commissioner of Canada.

Privacy officerprivacy@museai.ca · designation being formalized under Law 25
This page's history

Dated, because a page that
never changes is decoration.

October 1, 2026

Site measurement added. Google Analytics now loads alongside the Google Ads tag — after your yes, never before. Everyone who had already accepted is asked again, because that yes covered ads only.

September 29, 2026

Tracking, corrected. The site now asks before loading a Google Ads measurement tag, so “zero trackers” and “zero cookies” came off this page the same day — replaced by what is true: nothing loads without your yes.

July 29, 2026

First publication. Position stated; both signatures marked “signing in progress”; the ten principles mapped to mechanisms; the two automated behaviours disclosed; rights and privacy-officer contact published.