Beyond the AI Act: Europe’s Next Governance Challenge

Author: Denis Avetisyan

The landmark EU AI Act is just the first step, and a robust, independent agency is now vital to effectively manage the risks and ethical implications of rapidly advancing artificial intelligence.

This paper argues for the establishment of a strengthened supranational agency to complement the AI Act and ensure comprehensive AI governance, risk assessment, and ethical oversight within the European Union.

Despite the growing momentum surrounding artificial intelligence, a comprehensive and adaptive regulatory framework remains elusive. This paper, ‘From the AI Act to a European AI Agency: Completing the Union’s Regulatory Architecture’, revisits the question of optimal governance, arguing that while the EU AI Act represents a crucial initial step, a strengthened, independent supranational agency is essential for robust risk assessment, policy coherence, and the upholding of ethical principles. Such an agency would not only bolster the Union’s digital sovereignty but also facilitate effective international cooperation in this rapidly evolving technological landscape. Will a more robust institutional structure prove vital for navigating the complex challenges and opportunities presented by advanced AI systems?

The Emerging AI Landscape: Economic Promise and Systemic Risk

The artificial intelligence market is poised for explosive growth, with projections indicating a potential $15.7 trillion contribution to global GDP by 2030. This surge isn’t simply about technological advancement; it represents a fundamental reshaping of economic landscapes across sectors. While increased automation and enhanced productivity are anticipated benefits, this rapid expansion simultaneously introduces complex challenges. These include the need for workforce adaptation, potential exacerbation of existing inequalities, and the ethical considerations surrounding increasingly autonomous systems. Successfully navigating this period requires a proactive approach to harness AI’s economic potential while mitigating the risks inherent in such a transformative technology, ensuring broad-based prosperity rather than concentrated gains.

The projected surge in productivity driven by artificial intelligence, anticipated to reach 40% across sixteen key industries by 2035, is inextricably linked to a growing landscape of security vulnerabilities. As AI systems become increasingly integrated into critical infrastructure – from finance and healthcare to transportation and energy – the potential for malicious exploitation expands dramatically. These risks aren’t limited to conventional cybersecurity threats; they encompass data poisoning, adversarial attacks designed to manipulate AI decision-making, and the potential for autonomous systems to be hijacked or repurposed. The very sophistication that makes AI a powerful economic engine also creates complex attack surfaces, demanding a fundamental shift in security paradigms to proactively address these emerging risks and ensure the resilience of AI-driven systems.

The burgeoning global artificial intelligence market, projected to reach $2 trillion by 2030, is rapidly outpacing the capacity of current regulatory structures to address its complexities. Existing legal frameworks, largely designed for traditional industries, struggle to accommodate the unique challenges posed by AI’s autonomous nature, data dependence, and potential for unforeseen consequences. This regulatory gap creates vulnerabilities regarding accountability, bias, and security, demanding a shift towards proactive and adaptive governance. Simply reacting to issues as they arise proves insufficient; instead, policymakers must anticipate potential harms, establish clear ethical guidelines, and foster international cooperation to ensure responsible AI development and deployment, thereby maximizing its economic benefits while mitigating associated risks.

A Framework for AI Governance: The EU’s Pioneering Approach

The European Union’s AI Act establishes a legally binding framework for the regulation of artificial intelligence systems, representing a significant step toward proactive governance in the field. This legislation adopts a risk-based approach, categorizing AI applications based on their potential to cause harm. Systems deemed to pose an unacceptable risk – such as those manipulating human behavior or used for social scoring – are prohibited. High-risk systems, including those used in critical infrastructure, education, and law enforcement, are subject to stringent requirements regarding data quality, transparency, human oversight, and cybersecurity. The framework utilizes a tiered system of obligations, with lower-risk AI applications facing fewer regulatory demands, and minimal risk applications being largely unregulated. This structure aims to foster innovation while mitigating potential harms and ensuring fundamental rights are protected.

The practical application of the EU AI Act is largely dependent on the EU AI Office, established to oversee compliance and enforcement. However, this office currently operates with a constrained scope of authority and limited resources, prompting concerns regarding its ability to effectively monitor and penalize violations, particularly given the scale and rapid development within the AI sector. Critics point to the office’s reliance on national authorities for many enforcement actions, potentially leading to inconsistent application of the Act across member states. This structural limitation is particularly relevant considering the significant investment disparity, with the US receiving €62.5 billion in private AI funding in 2023, compared to the combined €9 billion in the EU and UK, potentially accelerating innovation outside the scope of EU regulatory oversight.

The EU’s AI regulatory strategy emphasizes Transparency, Accountability, and Data Privacy as core tenets for effective oversight. These principles are intended to ensure responsible AI development and deployment, but are challenged by significant disparities in private investment. In 2023, the United States attracted €62.5 billion in private AI investment, dwarfing the €7.3 billion received by China and the combined €9 billion for the European Union and the United Kingdom. This investment gap highlights a potential disadvantage for EU-based AI companies in terms of research, development, and scaling of AI technologies, necessitating a robust regulatory framework alongside strategies to attract and retain investment.

Beyond Oversight: Architecting a Dedicated AI Agency

The current structure of the EU AI Office faces limitations in proactive risk assessment and enforcement due to its reliance on existing regulatory frameworks and member state implementation. Agencification, the process of establishing a dedicated supranational agency, offers a potential solution by creating an independent entity with specific mandates and resources for AI oversight. This agency could operate beyond the constraints of national interests and harmonize standards across the European Union, allowing for more robust and consistent evaluation of AI systems and quicker responses to emerging risks. Such an agency would be empowered to independently assess AI technologies, enforce compliance, and promote responsible innovation, supplementing the work of existing national authorities and the EU AI Office.

A supranational AI agency, operating beyond reactive compliance, would conduct prospective risk assessments of AI systems before widespread deployment. This proactive approach extends beyond evaluating existing AI models to encompass the evaluation of proposed architectures, datasets, and intended applications. Enforcement of established standards would then be data-driven, focusing on identified risks and prioritizing interventions based on potential impact. Such a system aims to foster responsible innovation by preemptively addressing safety concerns, bias mitigation, and ethical considerations, thereby reducing the need for costly and disruptive corrective measures after deployment. This differs from current regulatory approaches which largely respond to incidents after they occur.

Technological sovereignty in the context of AI development refers to the capacity of a nation or bloc to independently control and oversee the full lifecycle of AI technologies, from research and development to deployment and maintenance. This includes establishing independent standards, conducting risk assessments, and enforcing regulations without undue reliance on external actors or technologies. Maintaining this control is considered crucial for safeguarding national interests, protecting data privacy, and ensuring that AI systems align with societal values and legal frameworks. The ability to exercise independent oversight mitigates potential vulnerabilities associated with dependence on foreign AI providers and promotes innovation within a secure and controlled environment.

The increasing adoption of Artificial Intelligence necessitates a blended regulatory approach utilizing both legally binding “hard law” and non-binding “soft law” instruments to foster international cooperation and adaptability. Current projections indicate a significant rise in AI implementation, with 78% of companies anticipated to be leveraging AI capabilities by October 2025, compared to 55% in 2023. Soft law, such as guidelines and codes of conduct, allows for rapid response to evolving technologies and facilitates broader participation beyond legally mandated requirements, while hard law provides the necessary enforcement mechanisms for critical risk areas. This dual approach aims to balance innovation with responsible development, accommodating the rapidly expanding AI landscape and encouraging consistent standards across borders.

Addressing Bias: Ensuring Equitable AI Outcomes for All

Artificial intelligence systems, while promising innovation, are susceptible to inheriting and even exacerbating pre-existing societal biases. These biases aren’t inherent to the technology itself, but rather stem from the data used to train the algorithms; if historical data reflects discriminatory practices – in areas like loan applications, hiring processes, or even criminal justice – the resulting AI will likely perpetuate those same inequities. Consequently, seemingly objective AI-driven technologies can systematically disadvantage certain groups, leading to unfair or discriminatory outcomes in critical life areas. This phenomenon isn’t limited to obvious biases; subtle patterns in data can also lead to unintended consequences, reinforcing existing power imbalances and creating new forms of digital discrimination. Addressing this requires diligent data curation, algorithmic transparency, and ongoing monitoring to ensure equitable results and prevent the automation of inequality.

The pursuit of artificial intelligence necessitates a steadfast commitment to fairness, a principle that extends beyond mere technical functionality. Achieving equitable outcomes demands rigorous attention throughout the entire AI lifecycle, beginning with data quality; biased or incomplete datasets inevitably lead to discriminatory results. Algorithm design itself must prioritize fairness metrics, actively mitigating potential biases embedded within the model’s logic. However, even technically sound algorithms require continuous outcome monitoring to detect and correct unforeseen disparities in real-world application. This proactive, multi-faceted approach – encompassing data, design, and diligent oversight – is not simply an ethical consideration, but a fundamental requirement for building trustworthy and beneficial AI systems that serve all members of society equitably.

The proliferation of AI-driven technologies – encompassing systems that generate human-like text, automate repetitive tasks, and ‘see’ and interpret the visual world – necessitates rigorous examination for potential discriminatory impacts. While promising increased efficiency and innovation, these tools aren’t inherently neutral; biases present in the training data or embedded within algorithmic design can lead to unfair or inequitable outcomes. For example, Natural Language Generation models might perpetuate harmful stereotypes through biased language, while Computer Vision systems could exhibit lower accuracy rates when identifying individuals from underrepresented demographic groups. Similarly, Robot Process Automation, if trained on biased historical data, may unfairly prioritize certain groups over others in decision-making processes. Consequently, proactive scrutiny, including diverse dataset creation, algorithmic auditing, and ongoing performance monitoring, is crucial to mitigate these risks and ensure these powerful technologies serve all segments of society equitably.

Despite the rapid integration of artificial intelligence – a strategy currently embraced by 83% of businesses – effective regulation is crucial to prevent the exacerbation of societal inequalities and safeguard vulnerable groups. Policymakers are increasingly focused on establishing frameworks that prioritize equitable outcomes, moving beyond simple compliance to actively assess and mitigate potential discriminatory impacts embedded within AI systems. This isn’t merely a matter of ethical consideration; ensuring fairness fosters trust and wider adoption, while proactively protecting those most susceptible to algorithmic bias-such as marginalized communities or individuals with limited access to technology-is becoming a central tenet of responsible AI governance. The challenge lies in creating adaptable regulations that keep pace with innovation, fostering a beneficial AI ecosystem while simultaneously upholding principles of justice and inclusivity.

A Vision for Responsible AI and Global Leadership

The rapid advancement of artificial intelligence necessitates a globally coordinated approach to ensure its responsible development and deployment, and a dedicated Supranational AI Agency is proposed as the central mechanism for achieving this. This agency would move beyond reactive regulation, proactively fostering innovation while simultaneously mitigating potential risks-such as bias, misuse, and unintended consequences. By centralizing expertise, standardizing ethical guidelines, and facilitating international collaboration, the agency would establish a unified framework for AI governance. This isn’t simply about controlling the technology, but rather about guiding its evolution to maximize benefits for all of humanity, encouraging a shared understanding of best practices and fostering a competitive landscape built on trust and accountability. Ultimately, such an agency is envisioned as a catalyst for progress, ensuring that AI serves as a force for good on a global scale.

The realization of artificial intelligence’s substantial economic benefits – a projected $15.7 trillion contribution to global GDP by 2030 – hinges critically on a foundation of Fairness, Transparency, and Accountability. These aren’t merely ethical considerations, but essential components for building trust and fostering widespread adoption of AI systems. Prioritizing fairness mitigates the risk of biased algorithms perpetuating societal inequalities, while transparency allows for scrutiny and correction of potential errors. Accountability establishes clear lines of responsibility, ensuring that developers and deployers are answerable for the impacts of their AI creations. By actively integrating these principles, the full potential of AI can be unlocked, driving economic growth and simultaneously delivering societal benefits that are both equitable and sustainable.

Technological sovereignty in artificial intelligence signifies a nation’s capacity to independently develop, deploy, and govern AI systems, ensuring alignment with its core values and strategic objectives. This isn’t simply about restricting access to foreign technologies, but rather fostering domestic innovation and expertise across the entire AI lifecycle – from data acquisition and algorithm design to infrastructure development and talent cultivation. By prioritizing self-reliance, a nation can mitigate risks associated with dependence on external actors, safeguard sensitive data, and proactively shape the ethical and societal implications of AI. This approach allows for the implementation of bespoke regulatory frameworks that reflect local priorities, rather than adopting externally imposed standards, ultimately fostering public trust and maximizing the benefits of AI for its citizens while preserving its unique cultural and political identity.

The establishment of a robust and internationally recognized AI governance framework is poised to solidify a leading position in the rapidly evolving technological landscape. This proactive approach isn’t merely about regulation; it’s about shaping the development and deployment of artificial intelligence to proactively address ethical concerns and maximize societal benefit. By championing principles of fairness, transparency, and accountability, this framework will serve as a blueprint for other nations, fostering global collaboration and ensuring AI technologies are aligned with human values. Such leadership will unlock the full economic potential of AI – estimated at $15.7 trillion to global GDP by 2030 – while simultaneously safeguarding against potential risks and promoting a future where AI genuinely serves humanity’s best interests.

The pursuit of robust AI governance, as detailed in the article, necessitates a holistic understanding of interconnected systems. This mirrors the sentiment expressed by Andrey Kolmogorov: “The most important things are the ones you don’t know.” The EU’s regulatory architecture, while progressing with the AI Act, demands continuous adaptation and foresight. Just as Kolmogorov highlights the significance of acknowledging unknowns, the article posits that a strengthened, independent agency is vital not merely to address known risks, but to proactively anticipate and mitigate the unforeseen consequences inherent in rapidly evolving AI technologies. A fragmented approach, focusing solely on immediate concerns, risks overlooking the broader systemic impacts-a principle central to both Kolmogorov’s insights and the article’s call for comprehensive AI governance.

Beyond the Algorithm: Charting a Course for AI Governance

The proposed EU AI Act represents a necessary, if belated, attempt to impose order on a swiftly evolving technological landscape. However, legislation alone, however comprehensive, is akin to detailing the components of a complex machine without understanding its operating system. A supranational agency, as this paper contends, is not merely a bureaucratic addition, but a critical component of the regulatory architecture itself-a central nervous system to interpret, adapt, and enforce the rules. The true challenge lies not in defining risk a priori, but in continuously reassessing it as the technology matures and its applications proliferate.

One cannot simply replace a faulty sensor without accounting for the entire feedback loop. The agency’s success will depend on its capacity to foster genuine interdisciplinarity, bridging the gap between technical expertise, legal frameworks, and ethical considerations. Crucially, it must avoid becoming a bottleneck, stifling innovation in the name of safety. The temptation to regulate based on present anxieties, rather than future possibilities, will be strong.

The coming years will reveal whether this proposed agency can truly function as a dynamic, responsive system – capable of anticipating, rather than merely reacting to, the unpredictable consequences of increasingly sophisticated artificial intelligence. The question is not whether the technology is controllable, but whether the regulatory structure can evolve at a comparable pace.

Original article: https://arxiv.org/pdf/2603.22912.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/