AI in Inspection Reports: Who Is Liable?

You are. The inspector who signs the report owns it, whether the first draft came from your fingers, a template, a findings bank or an AI assistant. There is currently no federal artificial intelligence law in force in Canada: what governs you is article 2100 of the Civil Code of Québec, the rules of your art — the BNQ 3009-500 standard — and your liability insurance. The tool changes your speed, not your standard.

Last updated: September 22, 2026

In this guide

Is there an AI law in Canada?
What the Civil Code actually says
Does AI change your level of liability?
What about Law 25?
Six safeguards
What to ask your software vendor
Quick questions

This article explains a public legal framework; it is not legal advice. For your own situation, talk to your insurer and a lawyer.

Is there a law governing AI in Canada?

Not right now. Federal Bill C-27 contained the Artificial Intelligence and Data Act (AIDA). It never got past committee stage in the House of Commons and died on the order paper when Parliament was prorogued in January 2025. As of this writing, no federal law governing artificial intelligence is in force in Canada.

That surprises a lot of inspectors, because the noise around AI suggests the opposite. But no AI law does not mean no framework. What applies to your report is what already applied to it:

What governs you	What it requires
Civil Code of Québec, art. 2098 to 2100	Your inspection contract is a contract of service: you must act with prudence and diligence, in accordance with the usages and rules of your art
BNQ 3009-500 standard	The rules of the art for pre-purchase inspection in Quebec: coverage, descriptive text, evidence, justification for anything not inspected
REIBH	From October 1, 2027: mandatory RBQ certificate, mandatory insurance, and compliance with BNQ 3009-500
Law 25 (private sector)	Protection of the personal information you collect and process, including inside your tools
Your contract and service agreement	The limits of your mandate, what you cover and what you exclude

What does the Civil Code actually say about your liability?

Article 2100 of the Civil Code of Québec is the sentence that decides, and it is worth reading word for word (original French):

« L'entrepreneur et le prestataire de services sont tenus d'agir au mieux des intérêts de leur client, avec prudence et diligence. Ils sont aussi tenus, suivant la nature de l'ouvrage à réaliser ou du service à fournir, d'agir conformément aux usages et règles de leur art, et de s'assurer, le cas échéant, que l'ouvrage réalisé ou le service fourni est conforme au contrat. »

Contractors and service providers must act in the best interests of their client, with prudence and diligence, and in accordance with the usages and rules of their art.

Three words carry everything: prudence, diligence, rules of the art. You are not required to see everything — an inspector owes an obligation of means, not of result. You are required to have acted the way a competent, prudent inspector would have acted in the same circumstances, following the recognized practices of the trade.

Notice what the article does not say: nothing about tools. Article 2099 even gives you explicitly "le libre choix des moyens d'exécution du contrat" — free choice of the means of performing the contract. Pencil, Word template, findings bank, drone, thermal camera, AI assistant: those are means. The law judges the result of your diligence, not the brand of your gear.

Does AI change your level of liability?

No. And that is the answer that puts most inspectors at ease once it is spelled out.

Think about what happened with earlier tools. When spell-check replaced manual proofreading, nobody argued the typo was no longer the author's responsibility. When thermal cameras arrived, they did not shift responsibility for interpretation onto the camera manufacturer. The AI assistant sits in the same family: it produces a draft, you produce a report.

What does change is the shape of the mistake you might make. With a generated draft, two specific risks appear, and both are manageable:

The plausible finding you never observed. An assistant can produce a perfectly credible sentence about something you did not look at. Writing in a signed report that a flashing is deformed when you never saw it is a fault — no matter who typed the sentence.
The review that falls asleep. When the text arrives clean, well structured and in the right vocabulary, the eye slides over it. That is the real danger, and it is human, not technological.

The countermeasure is easy to state and demanding to keep: never sign a sentence you have not approved. The distinction matters — it is not about writing every word yourself, or the assistant would be pointless. It is about reading, judging and owning: wording that comes out of a findings bank or a generated draft becomes yours the moment you approve it, exactly as a template agreement becomes your contract when you sign it. That approval, not authorship of the words, is what constitutes your diligence. That is the position we have defended at Lumos from day one, and it is why our writing assistant structures your dictations and photos into findings but delivers nothing without your approval, line by line. The platform also works entirely without AI — keyboard and findings bank — for inspectors who prefer to work that way.

What about Law 25?

This is the one place where a Quebec rule speaks directly to automated processing. Article 12.1 of the private sector act reads:

« Toute personne qui exploite une entreprise et qui utilise des renseignements personnels afin que soit rendue une décision fondée exclusivement sur un traitement automatisé de ceux-ci doit en informer la personne concernée au plus tard au moment où elle l'informe de cette décision. »

Anyone using personal information to make a decision based exclusively on automated processing must inform the person concerned.

The word that matters is exclusively. A report you review, adjust and approve before sending is not an exclusively automated decision — there is a human in the loop, and that human is you. The disclosure duty in article 12.1 does not target this situation.

Where Law 25 really concerns you is upstream: the personal information of your client — name, address, contact details, sometimes their circumstances — that you hand to your tools. Pasting a client's details into a consumer AI service is a communication of personal information to a third party. Always ask where that data lives and who can read it.

What safeguards should you put in place?

Six habits that cost a few minutes and hold up if your report is ever challenged.

Read every finding as though you had dictated it. Not skimming: out loud, for the important ones.
Check that every sentence maps to a real observation. If it is not in your photos or your notes, it does not go in the report.
Verify figures and references. A construction year, an insulation thickness, a clause of the standard: those are the details a generated draft can assert with confidence and that take ten seconds to check.
Keep your trail. Timestamped photos, field notes, dictations: that is what demonstrates your diligence, far more than which software you used.
Protect your client's information. Use professional tools where you know where the data lives, rather than a consumer service.
Talk to your insurer. A ten-minute conversation now beats a discovery at claim time. Ask in writing whether using drafting tools changes anything in your coverage.

What to ask your software vendor

The answers to these five questions say more than a page of marketing.

Question	Why it matters
Do I see and approve every line before it goes out?	It is the line between an assistant and an automaton
Does the tool work completely without AI?	Your practice should not depend on a function you do not control
Where is my data hosted, and my clients'?	A Law 25 obligation, and a matter of trust
Are my reports used to train models?	Your findings and photos are your asset, not raw material
What is the drafting based on?	A bank of verified findings and a standard, or a general model's memory?

To place the tools on the market against these criteria, our comparison of inspection software in Quebec puts the features side by side, and the pricing page gives our numbers with no quote to request.

Quick questions about AI and liability

Do I have to tell my client I use an AI assistant? No law requires it in this context today. Many inspectors mention it anyway, explaining that the tool structures their observations and that they validate every line. Framed well, it reads as rigour rather than a confession.

If the AI gets it wrong, can I turn to the software vendor? What you sign binds you to your client. Your relationship with your vendor is governed by your contract with them, and those contracts generally limit their liability. Counting on that is not a strategy.

Does my insurance cover a report written with AI? Your insurance covers your professional faults in the exercise of your duties as an inspector. The question to ask your insurer is not "do you cover AI" but "does anything in my policy change based on the tools I use". Get the answer in writing.

Does REIBH say anything about AI? No. The regulation governs inspector certification, insurance and compliance with BNQ 3009-500. It says nothing about tools. We covered its deadlines in our article on the real REIBH deadline.

Sources

Texts verified at the source on September 22, 2026:

AI in an inspection report: who is liable if the report is wrong?