On 16 September 2026, European Commission President Ursula von der Leyen said the Commission would invite leading frontier AI laboratories to discuss how the development of the most advanced models should be paced and made safer. She also proposed deeper cooperation with the United Kingdom, Canada and other partners on model evaluation, verification, early-warning systems and AI security.

This is a policy signal, not a new legal moratorium. No binding restriction on training frontier models was announced. The immediate legal framework remains the EU AI Act, including its obligations for providers of general-purpose AI models with systemic risk.

Existing legal duties

The AI Office has oversight powers in relation to general-purpose AI providers, including powers to request documentation, evaluate models and require corrective action.

The announcement suggests a policy direction towards more verifiable evaluation, effective safeguards and credible processes for responding to emerging capabilities. That observation is an inference from the speech, not a new statutory obligation.

  • Conduct and document model evaluations and adversarial testing.
  • Identify, assess and mitigate systemic risks.
  • Report qualifying serious incidents to the AI Office.
  • Maintain adequate cybersecurity protection for the model and its infrastructure.

Why downstream businesses should pay attention

Businesses integrating third-party models into customer-facing products should assess their own role, any modifications to the model, the intended use and whether the resulting system falls into a regulated category. Responsibility cannot simply be assigned to the model provider by assumption.

UK companies may fall within the EU AI Act’s territorial scope where they place an AI system or model on the EU market or the output of a system is used in the EU. The precise position depends on the facts and the Act’s scope provisions.

Practical steps

  1. Identify whether the organisation acts as a provider, deployer, importer, distributor or downstream modifier.
  2. Keep an inventory of models, suppliers, intended uses and territories of operation.
  3. Seek contractual access to relevant model documentation, evaluation results, incident notifications and material-update information.
  4. Test safeguards in the context of the actual product, rather than relying solely on general safety claims.
  5. Set escalation, suspension and incident-response procedures for unexpected model behaviour.
  6. Preserve evidence of risk assessments and governance decisions.

The policy direction is towards verifiable safety, continuing evaluation and accountability across the supply chain.

This article provides general information and does not constitute legal advice on a specific matter. Legal obligations depend on the facts and the applicable law.