Pilot action for the establishment of future Union Testing Facilities in AI

Topic description

ExpectedOutcome:

Deliverables

The main deliverables of the action should be:

• a tested and validated suite of testing procedures and tools to be used in a future full-scale UTF;
• training material for market surveillance authorities;
• recommendations for the long-term establishment of union testing facilities on AI.

In the context of the implementation of the AI Act, it is envisaged to establish Union Testing Facilities (UTF) in the area of artificial intelligence to support Member States authorities in the ex-post verification of the compliance of AI systems placed on the market with AI Act requirements whenever deemed necessary. Such Union Testing Facilities shall also provide independent technical or scientific advice at the request of the AI Board or market surveillance authorities. The main objective of this action is to launch a pilot project to prepare the implementation of the concept of testing facility for AI and define the requirements for a full-scale operation of the testing facility (after the pilot phase).

On the basis of the requirements of the EU Regulation on Artificial Intelligence (‘AI Act’) when adopted and the ongoing work aiming at developing corresponding standards (further to the standardisation request issued by the Commission and addressed to CEN-CENELEC), the action should perform the following tasks:

• developing an approach to the implementation of the UTF concept for AI systems, including notably the approach to the testing of the compliance of AI systems with the requirements introduced by the AI Act, as well as the expertise and infrastructure requirements of a future AI UTF;
• carrying out preliminary testing of AI systems (to validate the proposed approach);
• providing independent technical or scientific advice at the request of the Commission, the AI Board, or Member States market surveillance authorities;
• organising workshops and other activities fostering collaboration and exchange of information between Member States market surveillance authorities in preparation of the implementation of the AI Act;
• conducting initial and further training courses for the benefit of the staff of market surveillance authorities of Member States and the Commission,
• in the context of activities of standardization organizations, including notably CEN/CENELEC JTC 21, contributing to the development of standards to verify and validate AI systems against the AI Act’s requirements for High-Risk scenarios, whether stand alone or integrated in regulated products and services.