The European Commission is eager to expedite the integration of artificial intelligence (AI) within various industries while simultaneously collaborating with international partners to ensure a harmonious pace of development for advanced AI models, aimed at mitigating security risks. Thomas Regnier, spokesperson for the Commission on technological sovereignty, recently explained that discussions with leaders of significant AI laboratories are still in the planning stages, following an announcement made by Ursula von der Leyen during her State of the Union address.
The Commission distinguishes between the acceleration of AI usage in industrial applications and the coordination necessary for developing state-of-the-art models. The advancement of economic applications is contingent upon safety, adherence to European standards, and the preservation of fundamental rights. As mentioned by von der Leyen, a meeting with heads of AI laboratories is forthcoming; however, details regarding timing and specific agreements on the development pace remain undisclosed. The Commission defends its AI Regulation as a crucial tool for risk management, but specifics regarding the commitments companies will need to undertake or how international coordination will be structured have not been clarified.
In a press briefing, Regnier was asked whether the call to moderate the development of advanced models reflects the European executive’s policy. He clarified the Commission’s stance, emphasizing that while it advocates for accelerated AI applications in the economy, this must be balanced against the inherent risks associated with cutting-edge technologies. The Commission is keen on fostering an environment for rapid adoption within safe parameters, while also recognizing that the risks posed by the most capable AI models extend beyond the borders of the EU.
The industrial focus has already identified five priority areas, as outlined by von der Leyen in her speech on September 16. These sectors include health, transportation, the agri-food industry, advanced manufacturing, defense, and space. Von der Leyen also announced upcoming initiatives for November and highlighted the potential of European industries’ data as a resource for tailoring AI models to meet specific needs.
Regnier referred to recent incidents and warnings from AI company leaders pertaining to the advancement of sophisticated models, although he did not delve into specifics about these occurrences. He reiterated the Commission’s proposed course of action, which commences with collaboration between international partners, including the UK and Canada, and aims to facilitate discussions with AI laboratory heads. At the time of the briefing, preparations were still ongoing, with no definitive date set for the meeting.
Von der Leyen has also supported an international appeal, released by the Finnish Presidency, advocating for human oversight of AI systems. This document calls for pre-launch testing and independent evaluations with adequate access for assessors. It suggests that UN member states consider establishing an entity to set standards, allow audits, and convene nations when certain capacity thresholds of AI systems are exceeded. While the call presents a proposition for cooperation, it does not establish a formal institution.
When asked about how such an entity would fit into the existing framework and the roles of the US and China, Regnier asserted that Europe must take a leading role, though he did not specify the responsibilities of other collaborators. The Commission emphasizes that Europe’s regulatory experience could significantly contribute to an international response. Nevertheless, the briefing did not present a negotiated mandate or any proposed governance structure.
In the EU, the AI Regulation has already set specific obligations for providers of high-risk AI models, including evaluations, adversarial testing, and reporting serious incidents. These requirements apply specifically to this legal category and do not indiscriminately encompass all industrial AI applications.
Safety discussions occur against the backdrop of Europe’s current limitations in developing advanced AI models. According to expert conclusions published by the AI Office in July, urgent priorities include enhancing computing infrastructure and energy availability, alongside securing capital and attracting talent. The experts noted that the development of cutting-edge models is largely concentrated outside the EU. Their assessment, which does not reflect the official stance of the Commission, recommends bolstering Europe’s capacity to access, select, and control these technologies.
Regnier defended the adequacy of the AI Regulation, claiming that the European cybersecurity agency has access to the latest advanced models, yet did not provide evaluative results. The next significant step in relation to industry remains the anticipated meeting with laboratory leaders. However, the specifics of how development coordination will unfold and what commitments companies are willing to accept were not detailed in the briefing.
