What the Carrefour ruling actually decided

On 4 June 2026 the Tribunal judiciaire de Caen ordered Carrefour France to make carrefour.fr and its mobile app accessible within six months, under a penalty of 500 euros for each day it runs late. The part worth reading twice is not the penalty. It is what the court did with Carrefour's defence.

A conformance score is not a defence

Carrefour argued that its site was 71% conformant with the RGAA, the French accessibility reference framework. The court rejected that. Accessibility, it held, is an obligation of result: a shop cannot be partly accessible any more than a staircase can have most of a handrail. The applicable criteria have to be met.

If you sell a percentage as proof of compliance, or buy one, this is the sentence that matters. A score tells you where you stand and whether you are improving. On this reasoning it will not stand in for conformance.

What the decision does not say

Three things are worth being straight about, because the opposite is being written elsewhere.

The law is not settled. A similar claim against Auchan's e-commerce arm was rejected by the Lille tribunal on 5 May 2026, a month earlier. Two courts, two outcomes. Anyone presenting the Caen decision as the settled position is overstating it.
No regulator issued a fine. This came from disability associations, apiDV and Droit Pluriel with the support of Intérêt à Agir, going to court. The realistic trigger is a claim from people who cannot use your service, not an inspection.
It is a French decision. It applies the RGAA and the French transposition. The European Accessibility Act has applied across the EU since 28 June 2025 and each member state enforces its own version, so the direction of travel is shared while the mechanics are not.

What this means if you run a site

The uncomfortable part is that most of what got Carrefour here is not machine-detectable. Whether a blind customer can complete a purchase is a question about a journey, and no scanner answers it. Automated testing covers roughly a third of the WCAG success criteria: contrast, missing text alternatives, unlabelled controls, structure. Useful, and not the whole picture.

What automation is genuinely good for is finding the machine-checkable failures quickly, and leaving a dated record that you found them and fixed them. If a claim ever arrives, the argument you want to be able to make is not that you scored well. It is that you knew what was wrong and were demonstrably fixing it.

Start with what is findable

Conforma scans against WCAG 2.2 AA, gives a fix for every issue it finds, and keeps a dated record of the ones you close. It does not issue certificates, and it will tell you what it did not test.

Scan a page free

Common questions

Does a conformance percentage protect me?

Not on the reasoning in this decision. Carrefour put forward a 71% RGAA conformance rate and the court rejected it, holding that accessibility is an obligation of result: an online shop cannot be somewhat accessible, it has to meet the applicable criteria. A score is useful for knowing where you stand and for showing progress. It is not a defence.

Have regulators started fining companies under the EAA?

Not that has been made public. The pressure so far has come from disability associations bringing cases, not from market surveillance authorities issuing penalties. That matters for planning, because it means the trigger is a claim rather than an inspection, and claims tend to follow the sites people actually cannot use.

Is the case law settled?

No. A similar claim against Auchan's e-commerce arm was rejected by the Lille tribunal on 5 May 2026, a month before the Caen decision. One ruling in one direction and one in the other is not a settled position, and anyone telling you the law is now clear is selling something.

What does an automated scan actually cover?

Roughly a third of the WCAG success criteria: the machine-checkable ones such as contrast, missing text alternatives, unlabelled controls and page structure. Keyboard operation, whether a screen-reader user can actually complete a purchase, and whether the flow makes sense are not things a scanner can judge. Any tool claiming otherwise is overselling.

So what is worth doing?

Fix what is findable, keep a dated record of having done it, and get a human to walk the critical journeys. The record is the part people skip, and it is the part that demonstrates good faith if a claim ever arrives.

Sources

This page describes a public court decision and is not legal advice. Figures are limited to those the sources state.