With Power Comes Responsibility

Introduction

Today, artificial intelligence (AI) is already firmly embedded in our everyday lives. It analyzes our viewing habits to suggest tailored streaming recommendations or estimates arrival times by incorporating real-time traffic data. AI technology—in its best form—serves to make our lives simpler and impactful. In contrast, when a streaming service publicly called out 53 of its customers via social media for repetitively watching a particular movie, the users were made aware of the extent and power of user behavior tracing—and ultimately not amused. ¹ In every area where AI is deployed, the policy challenge remains to balance innovation with the protection of individual and societal rights. For biobanks, this is particularly critical. Accelerated access to large volumes of diverse, highly sensitive health, genomic, and lifestyle data from varied populations raises new questions for science and society. These challenges are amplified by the speed and power of AI systems, whose deployment does not occur in a vacuum. Moreover, the largely unchecked and uncritical integration of datasets across domains creates risks not only for individual donors but also for society as a whole.

Discussion

When performing the practice of biobanking, technological trends, cultural considerations, lobbying inputs from industry or patient advocacy as well as governmental priorities and economic considerations, among others, are rarely on the radar of what seems first and foremost a scientific and infrastructure effort. Biobanks are typically situated at the intersection of research and care. They often function as both service and research institutions, with their dominant role shaped by governance frameworks guided by broader national or regional science policy considerations. The U.K. Biobank, for instance, was established as an epidemiological cohort in the early 2000s in response to a strategic policy need to understand the genetic, environmental, and lifestyle determinants of common, complex diseases of U.K. citizens in middle and old age. ² Irrespective of the reasons for their establishment, biobanks have, over the past decades, accumulated extensive institutional knowledge in sensitive data management, integrating perspectives from multiple stakeholders across academia, industry, and diverse publics, including funders and patient advocacy groups. Moreover, biobanking and health data management have become an increasingly standardized and data-driven practice. It relies on fit-for-purpose large language models, which require at least community standards to fully realize the potential of large datasets.

While a lot has been accomplished toward digitalization from a technical perspective (e.g., basic machine readability of electronic health care records or informed consent forms), a lot remains to be accomplished when it concerns addressing societal implications. Consider, for instance, the efforts undertaken by scientists and policymakers to make data findable, accessible, interoperable, reusable (FAIR), with the aim of enabling more efficient data sharing, ³ This holds the promise for improved reproducibility, enhanced cross-domain research integration, and ultimately data valorization and commercialization. Despite extensive funding in this area, FAIR data are still not fully realized, perhaps because FAIR is often understood as a mere technical or policy tool rather than as something that requires a robust governance framework based on responsible data use that sustains trust over time.

Today, health data are collected for many reasons, ranging from scientific discovery and disease prevention to lifestyle management and self-care, making the health data industry one of the fastest-growing sectors globally. ^a In this context, AI development, to date, has been, for a long time, and in some sectors still is, a largely self-regulated industry. Although traditional legislative processes are always outpaced by innovation or societal progress, the pace with which AI has been evolving poses challenges beyond mere oversight and compliance aspects, and countries have taken different approaches to address this fundamental policy task. The European Union’s General Data Protection Regulation (GDPR) is considered the strongest and most influential privacy law. ⁴ Similarly, the EU’s AI Act ^b is argued as a comprehensive regulatory approach that aims to address AI through a risk-focused legal framework and is often criticized as bureaucratic and unfavorable to innovation. In contrast, the U.S. government emphasizes light regulation through sector-specific guidelines, which allow greater flexibility but lead to inconsistent and largely unenforceable rules across industries and states. Moreover, ethical considerations are often sidelined for the pursuit of economic growth and technological leadership. In contrast, California was among the first U.S. states that implemented very comprehensive regulations that signal toward a new direction in U.S. data privacy and consumer protection law. ⁵ China’s AI policies prioritize government surveillance and national security aspects that are closely tied to its economic strategies. At the same time, China has removed a comprehensive AI law from its 2025 agenda. Instead, it takes a flexible standard and risk management approach that is mindful of compliance costs and, in recent years, has positioned itself as a proponent of international AI rules.^6,7

This is just the tip of the iceberg showing the divergent policy priorities of just three major players. However, it already indicates that the uneven adoption and fragmented regulatory landscapes threaten to widen the gap between countries, societies, and economies. A global divergence of AI policies risks deepening economic, environmental, and societal inequality, increasing compliance costs, and leading to incompatibility of standards. When it comes to implications for science and society, however, significant policy and data gaps remain, including deficiencies in governance frameworks and stakeholder awareness. ⁸

Traditional consent models, for instance, are not well suited to AI-driven secondary data use, yet they continue to be applied due to time constraints and the lack of reconceptualization of a model that is rooted in 19th-century philosophy. Simultaneously, the use of AI in medical research is a paradox: AI is difficult to explain to participants due to its “black box” nature, yet it is increasingly utilized to render documents into machine-readable formats that improve data management. ⁹

Risks of algorithmic bias and inequitable representation are still poorly addressed, as awareness remains limited—often due to persistent data gaps, particularly affecting marginalized groups. Tensions between data sharing and privacy frameworks across countries remain conceptually underdeveloped and frequently manifest as systemic roadblocks, as, for instance, understaffed and overburdened legal departments tend to prioritize institutional liability protection over the facilitation of scientific progress.^10,11

Also, the concept of “data colonialism,” the appropriation of human life through massive data extraction by corporations and governments, transforming everyday behaviors into raw material for profit and social control, offers new ways to assess the implications of open access and fairness in data use. ¹²

New pathways, however, cannot be adequately achieved within the constraints of an individual research project, where solutions are often reactive and time-sensitive. Instead, addressing these implications requires careful reflection, long-term strategy, and structural change. The mantra “move fast and break things” is ill-suited to the domain of sensitive data management and biobanking, where public trust is essential and its loss may be irreversible. Trust-building, in this context, extends beyond a singular actor. It encompasses multiple stakeholders, including experts from academia–industry, policymakers, advocacy groups, and representatives of diverse publics, each with their own interests and expectations.

These interests, sometimes made explicit in strategy papers and policy statements, have profound implications: while they enable accelerated discovery and precision medicine, they require adequate measures to maintain public trust, enable meaningful ethical oversight, and permit equitable access to benefits. Addressing these issues requires revisiting and adapting core ethical principles, such as beneficence, nonmaleficence, and justice, alongside collective responsibility in ethical stewardship. ¹³ Here, biobanks have a critical role to play, especially because they operate at the intersection of biomedical research and health-related data services.

Let me elaborate. Safeguarding fundamental rights, human dignity, and personal choice has become increasingly challenging because data integration and access have not only intensified but have also become so complex that a one-off compliance assessment or consent procedure is insufficient. Such measures can uphold the promise of a robust governance system only briefly, until they are rendered obsolete by new discoveries. This can be addressed through adaptive, continuously updated governance frameworks that incorporate ethical review, stakeholder engagement, and iterative oversight mechanisms by design.

This can be achieved through structured engagement with stakeholders and continuous listening to societal concerns, including perspectives that extend beyond the purely medical domain, to ensure that governance frameworks remain responsive and inclusive. Put differently, the practice of biobanking needs to be framed within a broader societal discussion on data governance. Rather than relying on the “deficit model” of top-down, one-way communication, a new approach is needed that emphasizes two-way interaction between those who entrust, provide, and use data. ¹⁴ Such a feedback loop enables continuous learning about what is needed to maintain trustworthiness, transparency, and societal relevance in biobanking practices, as it incorporates AI. In this logic, compliance becomes a collaborative effort, much like scientific collaboration, which today relies on the interplay of interdisciplinary, intergenerational actors to tackle complex problems that affect generations and diverse environments.

Conclusion

New technologies, such as AI, operate at an accelerated speed and hold the potential for progress and economic growth. Their deployment requires resources, strategy, political will, and expert knowledge. At the same time, there is broad agreement across multiple sectors that AI tools are not perfect; they hallucinate and may generate biased solutions that can lead to undesirable results. Consequently, it is critical to retain accountability through human-in-the-loop or human-in-command approaches. ^c This, therefore, calls for adaptive governance approaches that are continuously updated in response to technological and scientific developments. Such approaches require the systematic integration of ethical and societal considerations into data governance, particularly in light of potential unintended or undesirable outcomes. Accordingly, a design-oriented perspective is needed that explicitly asks: Who stands to benefit and who bears the economic, societal, or environmental costs?

It is an all-society effort to reach a common agreement on what the power of AI can be used for and by whom. Biobanks, however, play a particularly important role not only due to their core function of managing highly sensitive health and genetic data but also because they can draw on decades of experience in translating bioethical principles into practical governance frameworks. For this, biobanks can rely on high-level ethical guidelines from international organizations, such as the UNESCO Recommendation on the Ethics of AI ^d as well as community-driven initiatives. ¹⁴ This positions biobanks as key actors at the intersection of biomedical research, data governance, and ethical oversight, where they are required to balance scientific progress with the protection of individual rights, public trust, and responsible data use.

Translating these recommendations into practical processes is, of course, challenging due to limited resources, the fast pace of technological development, and slow regulatory processes. This demands a collective responsibility across the global biobank community. Ultimately, any pathway to AI governance depends not only on international and transdisciplinary cooperation but also on the political will of institutions and individuals to commit to collaborative responsibility for the common good.