The European Union’s landmark Artificial Intelligence Act continues its staggered implementation, with new provisions activating this weekend, marking a crucial juncture one year after its initial enforcement. This ambitious legislative framework, designed to categorize and regulate AI systems based on their risk levels, aims to establish a global benchmark for responsible AI development and deployment. Its ongoing rollout highlights the complex challenge of governing a rapidly evolving technological landscape, prompting vital discussions on future adaptations and broader societal impacts.
Initiated on August 1, 2024, the EU AI Act meticulously defines AI systems into categories of unacceptable, high, limited, or minimal risk, each with corresponding prohibitions or conditions. While measures targeting unacceptable risk AI systems have been in effect since February 2, 2025, the current phase, activating on August 2, 2025, brings into force critical provisions related to robust AI governance standards, general purpose AI (GPAI) models, and a comprehensive sanctions regime. Despite this structured approach, full implementation of the extensive legislation is not anticipated until 2030, underscoring its long-term scope.
Despite its global aspirations, the act’s initial year has been characterized by a notable lack of widespread public impact, particularly concerning enforcement. Víctor Rodríguez, a senior lecturer at the Polytechnic University of Madrid, notes that without the activation of the sanctions regime, high-profile cases with “media impact” have yet to emerge. Furthermore, the desired “Brussels effect,” where global regulations follow the EU’s lead, appears uncertain, particularly with divisions among tech giants regarding the General Purpose AI Code of Practice, challenging the seamless adoption seen with previous European Union law initiatives like GDPR.
Nevertheless, subtle deterrent effects are already evident. Roger Segarra, an IT and intellectual property partner, points to the pre-emptive impact of prohibitions on practices like real-time remote biometric identification for surveillance, even before formal activation. A significant disparity in compliance readiness is also emerging; while larger corporations have proactively adjusted their AI systems and practices, small and medium-sized businesses (SMBs) face considerable bureaucratic and economic burdens due to limited practical guidelines and the sheer complexity of the artificial intelligence regulation. This uneven implementation is mirrored across member states, though Spain stands out for establishing its national AI supervisory authority.
The core challenge facing the AI governance framework is the relentless pace of artificial intelligence advancement. Arnau Roca, managing partner at Overstand Intelligence, emphasizes that AI’s evolution often outstrips legislative capacity, making continuous adaptation essential. He views the act as a necessary first step but highlights daily encounters with project requests that blur ethical boundaries, signaling the constant tension between innovation and regulatory oversight. This dynamic environment necessitates that technology policy remains agile and responsive to unforeseen applications and ethical dilemmas.
Experts also express concern over AI applications that either fall outside the current legal scope or pose unique existential risks. Beyond standard regulations, the potential for tools like real-time image recognition to become obsolete due to human ingenuity in imagining new uses is a significant worry. Rodríguez identifies “disruptive technological novelties” such as advanced AI agents, real-time deep fakes, and multimodal models as immediate concerns. More critically, the deployment of military AI systems on battlefields and the use of platforms like Palantir for global mass surveillance represent applications where the law’s reach is either absent or limited, raising profound questions about digital ethics and fundamental rights.
The ongoing evolution of intelligent technologies mandates that the European Union law be designed for continuous adaptation. Rodríguez anticipates adjustments within a few years, facilitated by the act’s structure, which prioritizes general principles while deferring technical specifics to harmonized international standards. This approach allows for more agile modifications to international standards compared to reopening the entire legislative process. However, the political and social pressures, as suggested by Segarra, could lead to a formal revision of the AI Act even before its projected five-year review period, emphasizing the need for robust a posteriori review mechanisms.
Despite its strengths, the act presents areas ripe for improvement. Concerns persist regarding its impact on open source AI systems, with ambiguities around the definition of “provider” creating uncertainty for projects like Llama and the regulation of third-party modifications. The administrative and compliance burden on smaller market players remains a significant hurdle, necessitating simplified procedures and subsidized tools for regulatory compliance. Furthermore, the clarity of practical exemptions is still lacking, and criticism surrounds the centralized database for high-risk systems, raising fears of trade secret leaks and concerns over an “excessive accumulation of power by Brussels” versus national-level oversight, challenging the very spirit of the technology policy.
