How will artificial intelligence standards and certifications be decided in Europe?

The definition of shared technical standards for the security of algorithms and other essential aspects related to the development of AI will be the main challenge related to the implementation of the AI Act, ultimately defining its true scope and ability to exert the much-desired control. These standards, however, take years to develop and approve (the average time to approve a standard is between three and five years). Much of the AI Act's prescriptive capacity is thus de facto postponed and subject to no small amount of uncertainty.

At the same time, these products will have to be certified and undergo an impact assessment with respect to the fundamental rights of these technologies. This will create a huge market related to the AI Act. In the absence of a detailed regulatory framework and, in the first instance at least, of these standards, a fundamental role will be played by those who are ready to intervene by offering a certification and assessment product that offers sufficient guarantees, including - if not only - with respect to possible liability profiles. An erroneous qualification of an AI system - for example, one that is wrongly not considered high-risk under Article 6 AIA or not identified as prohibited under Article 5 AIA - could expose the company to even significant penalties (7% of global turnover).

In the absence of a regulatory framework that has yet to be defined in its most operational (but essential) aspects, those who will be able to offer the market a convincing certification product and best practices will soon be able to establish themselves as leaders in this market. They will ultimately make a decisive contribution to defining the operational content of the AI Act and the criteria that companies wishing to innovate will have to adhere to.

It is likely that large global players that have been active for many years in the technological frontier and have developed internal best practices from data regulation will be able to offer these services sooner and better than others, and will soon dominate the market. Most likely, these will be non-European or, in any case, mainly non-European players that, with their solutions, will be able to significantly influence the real scope of application of what, in the intentions of the European Commission, should represent 'the European approach' to AI.

At the end of this process, however, we could perhaps ask ourselves how much of Europe will have been left and, similarly, we could assess whether the 'Brussels effect', on which the Commission clearly bet on, i.e. whether the AI Act model will have been able to influence the regulation of technology globally and beyond the borders of the member states, will have occurred.

*Andrea Bertolini, Associate Professor of Law, Sant'Anna of Pisa and Roberto Marseglia, CEO, DAAT