AFCO proposes human oversight of AI, election protection and global rules inspired by nuclear cooperation

Artificial intelligence used by European Union institutions should remain under human control, while Europe should develop both the capacity to protect elections and democratic processes and a stronger role in setting global AI rules. These directions are reflected in the compromises discussed by the European Parliament’s Committee on Constitutional Affairs, AFCO, in a report that also proposes exploring an international governance framework inspired by cooperation in the nuclear field. In brief: The report argues that human judgment and responsibility must remain central when EU institutions use AI. It treats deepfakes, disinformation, foreign interference and algorithmic manipulation as risks to elections and democracy. AFCO is discussing the creation of a parliamentary observatory on AI’s impact on institutions and democratic processes. The report proposes exploring a global AI framework inspired by international cooperation in the nuclear field, while Renew supports a treaty on the non-proliferation of high-risk AI systems. The text has not yet been adopted. The AFCO vote is scheduled for 1 October, while the final plenary vote is expected during the second October plenary session. The report is based on the idea that using AI in European institutions can bring concrete benefits, from faster legislative analysis and assisted drafting to more efficient multilingual cooperation. At the same time, the MEPs who negotiated the compromise insist that such tools must not assume responsibility for political, administrative or legal decisions. In the S&D intervention, Vitenys Povilas Andriukaitis argued that using AI in processes of constitutional relevance raises questions about where human judgment ends and automated intervention begins. He called for human oversight, a right to an explanation and the possibility of human review of automated decisions, so that institutions remain responsible for the outcomes of the systems they use. ECR supported the same basic idea and insisted that AI must not replace human oversight. The group linked human control to accountability and appeal mechanisms, as well as the protection of fundamental rights in situations where automated systems may cause discrimination, facilitate data or visual identity theft, or generate manipulated content. Election protection is one of the report’s central components. Rapporteur Emmanuel Kefalogiannis cited disinformation, deepfakes, foreign interference and manipulation as risks that AI can amplify in democratic processes. However, the text seeks to avoid portraying technology exclusively as a threat and argues that AI can also be used to strengthen institutions’ ability to identify and counter such risks. ECR insisted that citizens’ political views must not become targets of algorithms or AI-manipulated content and that the integrity of electoral processes must be protected against technological interference. This concern is one of the reasons behind the proposal to develop Parliament’s own capacity to analyse the effects of artificial intelligence. The report therefore retains the idea of a parliamentary AI observatory that would monitor the technology’s institutional and constitutional implications and its impact on democracy, decision-making processes and European integration. The rapporteur said that such a structure should not duplicate existing bodies, but argued that Parliament needs its own capacity to understand technological developments. The observatory proposal does not have unanimous support. Patriots for Europe welcomed the removal from the negotiated version of references to a possible amendment of the EU treaties and warned that a new structure could add bureaucracy or duplicate existing mechanisms. The group believes that priority should be given to implementing rules that have already been adopted. A stronger criticism came from ESN. Christine Anderson argued that the emergence of a new technology cannot in itself constitute a new competence for the Union and opposed the idea that a potential observatory could monitor political parties, elections or public discourse. She said AI could be used to support Parliament’s existing functions, but not to expand the EU’s institutional role. The report nevertheless goes beyond the internal functioning of institutions and proposes that the EU also promote an international AI governance architecture. Kefalogiannis supported exploring a framework inspired by the experience of international cooperation in the nuclear field, arguing that artificial intelligence cannot be managed effectively through national or regional rules alone. Under this approach, a future global framework should be based on common principles, transparency, safety and human control. The idea does not involve copying institutions built around nuclear technologies, but using their experience as a model for international coordination in a field whose effects transcend borders. Renew introduced a more specific proposal into this discussion. The group supports the EU promoting, at UN level, an international treaty on the non-proliferation of high-risk AI systems and capabilities considered particularly dangerous. The objective would be to prevent the uncontrolled development and use of advanced technologies and introduce common oversight and safety mechanisms. The global dimension is also linked to Europe’s technological position. The rapporteur warned that the EU cannot have real regulatory power unless it also has the capacity to develop the technologies it seeks to govern. He referred to the investment gap compared with the United States and presented investments in computing capacity, infrastructure and AI gigafactories as strategic projects for competitiveness and technological sovereignty. The debate also argued that dependence on data centres, platforms or systems developed in other jurisdictions may limit the Union’s regulatory autonomy. The report thus seeks to link democratic control to domestic technological capacity and Europe’s position in the global AI competition. The social dimension completes this approach. Kefalogiannis argued that AI must support and strengthen workers’ capabilities and that economic transformation must be accompanied by investment in skills and education. The report rejects the idea that responsibility for the effects of AI systems should be transferred to users, consumers or employees. In this part of the debate, Clara Dobrev of S&D went further, warning that AI development controlled mainly by large companies could concentrate both wealth and political power, accelerate the loss of jobs and affect social stability. This is, however, her political position expressed during the debate, not a common conclusion of the report. The report discussed by AFCO does not yet represent the position of the European Parliament. The compromises were negotiated between the rapporteur and shadow rapporteurs, and the committee vote is scheduled for 1 October. If the text is adopted, the final vote in Parliament is expected during the second October plenary session. The most ambitious elements, including the parliamentary observatory and the idea of an international treaty for high-risk AI systems, therefore remain proposals in the parliamentary process and are not decisions already endorsed by the European Union.,