EU and Canada plan mutual recognition of AI assessments

This analysis first appeared as the weekly Praxikon newsletter.

EU and Canada work towards recognition of AI assessments

On 30 October, the EU and Canada agreed to work towards mutual recognition of conformity assessment results for AI systems. These assessments examine whether a system meets specified requirements. The intention appears in paragraph 26 of their joint statement, following the summit in Montréal on 29 and 30 October.

The Council’s summary places this alongside cooperation on AI security and governance. Our reading for your procurement records: an agreement to work towards recognition is not, by itself, evidence that a particular foreign certificate satisfies the applicable European requirements.

What does this mean for you?

Do you buy AI from a supplier abroad, or provide AI across borders yourself? When a supplier makes a compliance claim, ask for the underlying assessment. Which requirements were assessed, by whom, for which system version and for which use? A quality mark without that information does little to support your own decision.

Our practical recommendation is to distinguish three things: a policy intention, an assessment of a specific system and the arrangements for your own use. For example, a test of an earlier version is not automatically evidence for the version you now plan to deploy. Record the difference and any unanswered questions.

The transparency date does not move

The transparency obligations in Article 50 have applied to systems within their scope since 2 August 2026, under Article 113. There is a limited transition period: providers of generative AI systems placed on the market before 2 August 2026 must comply with the machine-readable marking and detectability requirements of Article 50(2) by 2 December 2026. This follows from Article 111(4).

That transition period concerns only this provider obligation. It does not postpone the separate duty on deployers to disclose deepfakes under Article 50(4).

Action for this week

Choose one AI application for which you will soon renew a contract or introduce a new version. Review these three points with the process owner and procurement:

1. The claim. Keep the supplier’s exact statement and ask which document supports it. Record whether it concerns a standard, a test result, a certificate or an announced arrangement.
2. The fit. Compare the system version, intended use and assessed requirements with your own application. Ask specific follow-up questions about each difference.
3. The follow-up. Record who will request missing evidence and when you will review it again. If you use generative AI, also ask how the supplier implements the marking required by Article 50(2).

This is our practical recommendation, not a legally prescribed checklist. The result is one substantiated procurement decision, with open questions clearly recorded.

Would you like to strengthen careful AI use within your team? Explore role-based AI literacy training from LearnWize. Training supports people’s actions; it does not replace an assessment of an AI system.

Until next Tuesday.
Zahed Ashkara
Praxikon