Abstract
INTRODUCTION
In the span of the preceding half-year, artificial intelligence (AI) has attained a remarkable surge in prominence, evolving into an entity impossible to disregard. The user-friendliness inherent in these nascent tools, exemplified by AI-propelled text and image generators, has catalyzed substantial discourse concerning the judicious application of AI technology. Concurrently, legislative bodies, such as Congress, have embarked on an exploration of AI governance. Deliberations have gravitated towards an expansive spectrum of societal ramifications stemming from AI, encompassing even the realm of biosecurity hazards. 1
AI’s presence in the field of life sciences has a venerable history, dating back to the foundational concepts introduced by Alan Turing in the 1950s. These ideas blossomed into practical applications by bioinformaticians, who use AI in genome analysis, long before the dawn of the 21st century. 2 The U.S. Congress has taken its initial steps in addressing AI-associated risks, focusing primarily on societal impacts. 3 Issues like the spread of misinformation, automated employment determinations, and other concerns affecting the general populace have become the forefront of legislative deliberations. 4 These concerns are not only wise but also resonate globally.
Although recent legislative proposals have aimed to address AI-related issues in the health care sector, there has been limited attention given to crafting solutions tailored to mitigate risks arising from the convergence of AI with the biosciences and biotechnology sectors. Safeguarding these tools requires a meticulous approach, an understanding of their operational intricacies, and the potential for misuse. Central questions regarding the oversight of AI in biosciences revolve around strategies to supervise AI-powered bio-centric tools such as AlphaFold2 5 by Google DeepMind.
Google DeepMind has introduced an advanced AI system known as AlphaMissense, designed to predict DNA variations that may lead to diseases. 6 This technology could revolutionize the diagnosis of rare disorders and offer valuable insights into drug development. AlphaFold, another DeepMind creation, significantly advanced the accurate prediction of protein structures, a monumental achievement in biology. 7 However, the company has chosen not to release the entire model for unrestricted use due to biosecurity concerns. The intent is to mitigate biosecurity risks while promoting innovative research and development. This raises the question of whether alternative methods can be devised to prevent these technologies from being misappropriated. 8
The U.S. Bioterrorism Act of 2002 stands as a crucial milestone in biosecurity regulation, 9 aiming to enhance the nation’s capacity to prevent and respond to bioterrorism and public health crises. 10 Unlike its predecessors, which focused on punitive measures after incidents, 11 this legislation proactively identifies potential biological threats to animal, plant health, and public safety. 12 The United States Department of Agriculture (USDA) and the Food and Drug Administration (FDA) play pivotal roles in this endeavor.
Before the U.S. Bioterrorism Act of 2002, the Biological Weapons Anti-Terrorism Act of 1989 criminalized the use of biological agents as weapons but allowed their use for peaceful purposes. 13 Subsequently, the Patriot Act criminalized the possession of these agents for nonpeaceful purposes. 14 Section 212 of the U.S. Bioterrorism Act of 2002 now governs the regulations for biological agents, striking a balance between research facilitation and threat mitigation. 15
Various legislative efforts have sought to initiate thorough investigations, establish new regulatory bodies, or outline comprehensive frameworks for risk management, exemplified by the AI Risk Management Framework under the National Institute of Standards and Technology (NIST). 16 Recently, the Biden Administration has engaged with AI model pioneers, aiming to implement measures that mitigate risks, including those related to biosecurity. 17 It is crucial to note that the current supervisory mechanisms for AI models are primarily voluntary, sparking discussions about providing incentives and considering more potent regulatory approaches. As AI models continue to proliferate and those dedicated to biosciences and biotechnology become more sophisticated, the need for enforceable regulations and accountability thresholds becomes increasingly urgent.
THE CONFLUENCE OF AI AND BIOSECURITY: CONCERNS AT HAND
In the rapidly evolving landscape of AI and its intersection with the field of biology, the acquisition and utilization of extensive biological datasets emerge as fundamental prerequisites for the development of AI applications tailored to the realm of life sciences. These datasets, encompassing a broad spectrum of biological information from intricate protein structures to comprehensive genomic sequences, constitute the bedrock upon which AI tools for biological applications are constructed. 18 However, as we delve further into this amalgamation of AI and biology, we are compelled to address a pressing issue of paramount significance. It calls for a meticulous equilibrium to be struck between facilitating scientific progress and mitigating the latent threats to public health and national security posed by the potential misuse of these very datasets.
AI, at the vanguard of technological innovation, is poised to bring about revolutionary changes across multiple domains of human existence. Its potential is nothing short of breathtaking, holding the promise of breakthroughs in an array of disciplines, from education to biology. AI models, notably exemplified by the likes of ChatGPT, have already transformed the landscape of learning and information dissemination. These models are not mere tools; they are catalytic agents of change that have engendered remarkable achievements. These milestones encompass solutions to intricate challenges, such as the enigmatic protein folding problem and the expeditious acceleration of drug discovery. 19 These feats, among numerous others, have the potential to redefine the boundaries of scientific knowledge and health enhancement. However, this tide of progress is accompanied by lurking shadows of peril.
Although AI propels advances in biology and democratizes access to knowledge, it concurrently unfurls concerns regarding its potential for malevolent exploitation. The very technology that empowers scientists, students, and enthusiasts to delve into the intricacies of biology also furnishes malevolent actors with tools that may yield catastrophic outcomes, akin to the fearsome specter of Aum Shinrikyo. 20 This fundamental issue has been underscored in recent research, unveiling the elevated risks associated with biological weapons and bioterrorism, all made more potent by expansive language models [large language models (LLMs)] such as ChatGPT and novel AI-driven tools tailored for biological manipulation. A study published in the Nature Machine Intelligence journal in 2019 documented that the conceivable utilization of AI in the design of chemical weaponry presents an imminent threat. 21 The study proffers an unsettling observation: AI can expedite the creation of highly toxic chemical compounds. Another example lies in the work of scholars from the Massachusetts Institute of Technology, who have unveiled the capacity of LLMs within Chatbot frameworks to facilitate the concoction of viruses with pandemic potential, even by individuals devoid of laboratory training. 22
Expansive language models, typified by the noteworthy ChatGPT, are pivotal in enhancing accessibility to knowledge concerning biological weapons. While they possess inherent limitations in imparting tacit or background knowledge essential for bioweapon development, there exists a tangible risk that these models can bridge this knowledge gap, thus potentially ushering in an era of increased experimentation aimed at sinister outcomes. 23 Moreover, as AI technologies advance, language models are destined to play an even-more-integral role in guiding laboratory automation and strategic experiment design, making regulation and ethical oversight all the more critical to prevent their abuse. 24
Specialized AI tools, exemplified by biological design tools (BDTs) such as AlphaFold2 and RFdiffusion, leverage biological datasets to revolutionize numerous facets of biological engineering. 25 These tools, while promising remarkable potential for scientific advancement, concurrently introduce apprehensions related to biosecurity. The concerns span the potential creation of bioagents endowed with unprecedented attributes, transcending the evolutionary constraints governing naturally occurring pathogens. This raises the unsettling possibility that malevolent entities may develop bioagents far more destructive than their natural counterparts, fundamentally transforming pandemics from disasters into existential crises. 26 These tools also cast a foreboding shadow over the prospect of bioagents tailored to specific targets, be they geographical or demographic in nature.
In recognition of the gravity of these concerns, apprehension has crystallized among policy makers, transcending political boundaries. A Senate Judiciary Committee subcommittee hearing recently convened to delve into the malevolent potential of AI, particularly concerning the construction of biological weaponry. 27 This concern has also resonated within the industry, with prominent entities like OpenAI, Alphabet, and Meta Platforms voluntarily pledging to implement security-oriented measures aimed at bolstering the safety of AI technologies. 28 Such measures include the watermarking of AI-generated content. Policy makers globally are now increasingly focused on devising strategies to mitigate the perils posed by the proliferation of AI technologies, mindful of the overarching implications for national security and economic stability. 29
In an era where the ascension of AI transcends political boundaries, the preservation of the nation’s health security emerges as an imperative of paramount national importance, irrespective of partisan divisions. This era confronts us with the tangible specter of AI-forged perils, encompassing the potential emergence of threats on par with bio-chemical armaments. Eric Schmidt, the former Chief Executive Officer of Google, has sounded a clarion call regarding AI’s potential complicity in engendering biological strife, especially as the accessibility to viral databases proliferates. 30 Empirical endeavors have incontrovertibly demonstrated AI’s capacity to amplify the virulence of viruses, molecules, and deleterious bacteria, thereby enhancing the lethality of infectious or other harmful agents. In the quest for a prudent resolution, it is essential for nations to establish a consensus on a cohesive framework of regulations governing AI’s role concerning bioweapons. 31 Foremost among these regulations should be an unequivocal prohibition against employing AI to inflict harm upon human lives. 32
Biological armaments, comprising microorganisms or pernicious substances derived from living entities, possess the potential to spread on an unprecedented scale within human populations. 33 They present formidable challenges to public health, inciting rapid-spreading diseases and epidemics that result in enduring incapacitation and widespread casualties. Historical instances of the deployment of biological weaponry underscore the cataclysmic potential inherent in these agents, emphasizing the necessity for stringent regulatory protocols. 34 The realm of bioterrorism presents a complex enigma, with the ethical guardianship of biological initiatives falling upon sovereign states. The capability of AI to generate malevolent microorganisms has been empirically substantiated, as researchers have succeeded in producing a panoply of deleterious agents within a mere six-hour span. 35 Thus, the regulation of AI’s scope becomes a pivotal defense against facile biotic agent production.
Strengthening the foundation of biosecurity and reinforcing public health infrastructures is of utmost importance. 36 This mandates the development of effective methods for waste disposal, rigorous accountability in waste tracing, comprehensive training for health care personnel, collaborative coordination with cyber-terrorism entities, and the establishment of a dedicated oversight body to supervise AI developmental endeavors in the domain of biological commodities. Governments and public health authorities must fully acknowledge the perils that converge at the intersection of AI and bioweapons. 37 To forestall the unbridled maturation of AI capabilities, stringent regulatory frameworks emerge as an absolute necessity. Confronting these impending threats demands unwavering determination and prolonged steadfastness, transcending political divisions for the greater good of national and global security.
PRESIDENT BIDEN’S PLEDGE TO HARNESS AI’S POTENTIAL
Since taking office, President Biden, in collaboration with Vice President Harris and the entirety of the Biden-Harris Administration, has been steadfast in harnessing the vast potential of AI while vigilantly addressing the associated risks. In doing so, the administration prioritizes safeguarding the rights and safety of the American populace. 38 A testament to this commitment is President Biden’s orchestration of a gathering of seven prominent AI enterprises at the White House, including Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI. 39 This gathering serves as the platform for the Biden-Harris Administration to announce voluntary commitments from these corporate entities, signifying a collective effort toward the responsible, secure, and transparent advancement of AI technology.
In the endeavor to fully unlock AI’s potential, the Biden-Harris Administration is leading the charge in establishing exacting standards to ensure that progress does not compromise the rights and safety of the American citizenry. 40 The commitments made by these corporations underscore three fundamental principles that must serve as the foundation of AI’s future: safety, security, and trust. 41 These commitments represent a significant stride toward the cultivation of conscientious AI. As the pace of innovation continues to accelerate, the administration remains resolute in reminding these corporate entities of their obligations and taking decisive measures to ensure the security of the American populace.
Currently, the Biden-Harris Administration is in the process of crafting an executive order and pursuing bipartisan legislation, both of which are designed to position America at the forefront of responsible innovation. The seven leading AI conglomerates have made the following pledges:
42
Assure Product Safety Before Public Introduction: These companies commit to comprehensive internal and external security evaluations of their AI systems before public release. These evaluations, including input from external experts, serve as vital safeguards against critical AI risks such as biosecurity, cybersecurity, and broader societal implications. They also pledge to foster information exchange within the industry and with various stakeholders, including governments, civil society, and academia, to effectively manage AI risks. This collaboration includes sharing best practices for safety, insights into attempts to circumvent protective measures, and avenues for technical collaboration. Prioritize Security in System Development: The corporations commit to substantial investments in cybersecurity measures and safeguards against insider threats to protect proprietary and unreleased model weights. Recognizing that these model weights constitute the core of an AI system, the companies collectively emphasize their release only under secure circumstances and intentions. They also pledge to facilitate third-party identification and reporting of vulnerabilities within their AI systems, thereby expediting the identification and resolution of issues post system release. Earn the Confidence of the Public: The companies commit to establishing robust technical protocols to indicate when content is AI-generated, facilitated by the implementation of a watermarking system. This measure encourages creative AI applications while curbing the perils of deception and fraud. Furthermore, the companies pledge to publicly disclose their AI systems’ capabilities, limitations, and domains of appropriate and inappropriate application. This comprehensive report will address both security and societal concerns, including facets such as fairness and bias. Engage in Societal Safeguarding Research: The corporations commit to prioritizing research aimed at mitigating the societal risks posed by AI systems, including efforts to counter harmful bias and discrimination while preserving privacy. Recognizing the historical prevalence of these hazards within the AI landscape, these companies pledge to deploy AI systems that effectively mitigate such dangers. They also commit to developing and deploying advanced AI systems dedicated to addressing humanity’s most pressing challenges, ranging from cancer prevention to climate change mitigation, with a focus on prudent management that can greatly enhance prosperity, equality, and security for all.
LEGISLATIVE RESPONSE: U.S. AI-BIOSECURITY INITIATIVES
Senators Edward Markey of Massachusetts and Ted Budd of North Carolina have jointly introduced a pair of bipartisan legislative measures, aimed at empowering the Federal government to comprehensively understand and address potential threats to public health security arising from the rapid evolution of AI technologies. 43 These legislative initiatives, known as the “Artificial Intelligence and Biosecurity Risk Assessment Act” 44 and the “Strategy for Public Health Preparedness and Response to Artificial Intelligence Threats Act,” 45 signify a pivotal step toward enhancing the nation’s readiness to confront biosecurity risks facilitated by AI.
This legislation calls upon the Department of Health and Human Services (HHS) to assume a more proactive and substantive role in assessing the implications of AI and its associated risks to biosecurity. Specifically, the Assistant Secretary for Preparedness and Response within HHS is entrusted with the solemn duty of conducting meticulous assessments of the risks associated with the potential exploitation of advanced AI technologies in the creation of various categories of biosecurity threats. These encompass uncharted pathogens, viruses, bioweapons, or even chemical agents of destruction. 46
This complementary legislative measure broadens the responsibilities of the HHS by mandating the agency to develop a comprehensive strategy for countering and mitigating the risks posed by AI to national health security, particularly through the avenues of biological and chemical weaponry. 47 The strategy is to be meticulously formulated and presented to Congress within 180 days of the enactment of the legislative bill. 48
A pivotal aspect of this legislative initiative involves an amendment to Section 2811 of the Public Health Service Act (42 U.S.C. 300hh-10). This amendment introduces a new subsection denoted as “(h) ASSESSMENT OF EMERGING RISKS.” Within this context, the Assistant Secretary for Preparedness and Response is tasked with the responsibility of conducting risk assessments and operationalizing strategic endeavors to investigate whether the evolution of AI technology, including open-source AI models and expansive language models, could be harnessed, either deliberately or inadvertently, to create unprecedented forms of pathogens, viruses, bio-agents, or toxic substances. The scope of these strategic actions may include:
49
Diligently monitoring and investigating potential global catastrophic biological risks where biological agents could catalyze sudden and cataclysmic loss of human lives, with enduring ramifications on national governments, international diplomatic relations, economies, social equilibrium, or the global security framework. Incorporating the outcomes of risk assessments from this subsection into the overarching National Health Security Strategy, as delineated within Section 2802.
In addition, the strategic framework devised pursuant to the legislation outlines a comprehensive protocol for public health preparedness and responsiveness, with the aim of mitigating the latent risks posed by the misuse of AI, which could threaten national health security. This framework delineates responsibilities, roles, and preparedness benchmarks, attributing them to the office of the Secretary. The overarching objective is to enhance the nation’s capacity to anticipate, respond, and mitigate the perils emerging from AI-related threats. It includes the development of metrics to gauge the attainment of preparedness milestones and the effectiveness of risk mitigation efforts. 50
Moreover, the legislative response aims to identify and address gaps within the spectrum of public health capabilities, which are crucial for achieving preparedness benchmarks and risk reduction objectives.
51
Strategies are carefully devised to rectify these gaps and enhance collective capabilities for managing public health crises, particularly in the context of emerging AI-related threats. The strategic framework also addresses challenges posed by the malicious use of AI, including:
52
A comprehensive assessment and mitigation strategy for the perils posed by the deployment of AI in the creation of biological weaponry. Strategies to counteract the design and crafting of viruses and bacteria that exhibit resistance against established modes of treatment. A commitment to mitigating other latent hazards originating from the hostile manipulation of AI, as determined by the Secretary.
SENATE DISCUSSIONS: CHARTING THE PATH FOR AI REGULATION AND OVERSIGHT
On September 12, 2023, the Senate Energy Committee convened a pivotal hearing focused on the recent advancements in AI and their far-reaching implications for U.S. technological competitiveness. 53 This gathering featured prominent contributions from Microsoft and chipmaker Nvidia, who offered insights on the regulatory aspects of AI. Senator Richard Blumenthal played a key role by advocating for a risk-based approach to AI regulation, and he, along with Senator Josh Hawley, a Republican from Missouri, introduced a bipartisan AI framework. 54 This framework recommends that companies engaged in AI development should register with an independent oversight body responsible for licensing high-risk AI technologies, although the full text of this proposal is not yet publicly available. Furthermore, it calls upon Congress to address a specific section of the 1996 Communications Decency Act that currently leaves tech companies vulnerable to liability and potential lawsuits when developing AI tools. 55
The call for external oversight and regulation
Bianca Recto, communications director for Accountable Tech, echoed a clear sentiment during the hearing. Recto emphasized that the track record of big tech companies in terms of self-regulation has often aligned with their self-interest, underscoring the pressing need for external oversight. 56 Both Microsoft and Nvidia, in their opening testimonies, expressed support for the Senate’s efforts to create a legal framework for certifying “high-risk” AI through an oversight board. Importantly, they emphasized the distinction between advanced AI systems and less capable ones. However, Neil Hartzog, an expert in privacy and technology law, cautioned Congress against settling for half measures and industry-led approaches. 57 He urged the inclusion of mechanisms for enforcing liability and other critical regulatory safeguards, in addition to encouraging transparency, mitigating bias, and promoting ethical principles.
Microsoft President Brad Smith specifically called for the implementation of a “safety brake” for AI systems responsible for managing critical infrastructure, such as power grids and water systems. 58 The hearing explored various facets of AI’s applications, including its use on battlefields, in medical laboratories, and for disaster response and security threats. Moreover, the committee scrutinized the growing risks that AI poses to safety, privacy, civil rights, democracy, and the broader economy.
CASE STUDY: AI IN ANTIMICROBIAL RESISTANCE
The development of AI has ushered in a promising era in the realm of antimicrobial resistance (AMR) management. 59 AI has made significant strides in recent years, enhancing our ability to combat antibiotic-resistant microbes, marking a pivotal advancement in public health. 60 AI’s role in AMR begins with its profound impact on data collection and preparation. Through sophisticated data analysis and machine learning algorithms, AI systems can efficiently process vast datasets, identifying critical patterns and insights that would be challenging to discern through conventional means. 61 This capability lays the foundation for the deployment of AI-driven systems that can revolutionize our approach to AMR management.
Within this landscape, the “Strategy for Public Health Preparedness and Response to Artificial Intelligence Threats Act” assumes a critical role. This legislation is expressly designed to address the intersection of AI and public health, including its application in managing AMR. It imposes a set of legal obligations aimed at safeguarding national health security in the face of potential risks arising from AI technologies. 62 Under this Act, agencies are obligated to formulate a comprehensive strategy geared towards countering and mitigating the specific risks that AI may pose to public health, with a particular focus on biological and chemical threats. 63 These risks encompass the possibility that AI, including open-source AI models and expansive language models, could be harnessed intentionally or unintentionally to create novel pathogens, viruses, bio-agents, or toxic substances that could pose grave threats to health security. 64
One of the central mandates of the Act is the requirement for thorough risk assessments. Agencies are tasked with systematically evaluating the potential hazards associated with the evolution of AI technology. 65 This involves scrutinizing the capabilities and potential misuse of AI systems to forge pathogens or agents resistant to established treatments. By conducting these assessments, the Act aims to proactively identify and counteract any design or crafting of viruses and bacteria that exhibit resistance to existing modes of treatment.
CRITICAL ANALYSIS
The health protection legislation globally has an illustrious history rooted in international legal instruments, beginning with foundational documents like the Universal Declaration of Human Rights. 66 These instruments, such as the International Covenant on Economic, Social and Cultural Rights, the European Convention on Human Rights, Convention 108+, and the European Social Charter, house general provisions pertaining to health protection and related rights. 67 Yet, for the effective regulation of AI in the health care domain, these general instruments require further refinement and adaptation.
Sector-specific contextualization of health protection principles can be found in the Universal Declaration on Bioethics and Human Rights by the United Nations Educational, Scientific and Cultural Organization (UNESCO) and the Oviedo Convention by the Council of Europe. 68 The Oviedo Convention, uniquely, is the sole multilateral binding instrument entirely dedicated to biomedicine. However, these principles must be broadened and adapted to effectively regulate AI in health care. Complementing the Convention are two nonbinding instruments: the Recommendation on health data and the Recommendation on research on biological materials of human origin. Both underscore the interplay between biomedicine, health care, and data processing. 69
Although these international instruments were conceived in a pre-AI era, they offer critical safeguards related to self-determination, human genome treatments, and research involving human subjects, which remain pertinent even with the emergence of AI applications in health care. 70 However, the concept of self-determination encounters similar challenges in both biomedicine and data processing. The intricate and often opaque nature of AI applications can obstruct individual autonomy in decision-making, whether it pertains to medical treatment or data processing.
Binding international instruments in the field of biomedicine contributes significantly to the foundational principles and values that can be extracted from them. These principles provide a crucial foundation for the development of future AI regulations in health care. 71 Key principles derived from these instruments encompass areas such as the primacy of the human being, equitable access, acceptability, the principle of beneficence, private life and the right to information, professional standards, nondiscrimination, the role of experts, and public debate. 72 These principles serve as building blocks for shaping comprehensive AI regulation that effectively addresses the intricate challenges posed by AI in the field of health care and biosecurity.
The foundation of an international expert Working Group, led by the Nuclear Threat Initiative and the World Economic Forum, signifies a pivotal moment in the quest for global biosecurity norms. 73 This group aims to construct a robust framework to effectively deter the misuse of synthetic DNA. The International Gene Synthesis Consortium, established in 2009, is a cornerstone in the global biosecurity architecture. 74 Its primary objective is to harmonize screening protocols among its member companies, striking a delicate balance between security measures and potential risks. Transparency is crucial in both the development and testing of this common mechanism to maintain international trust and legitimacy. 75
Striking the right balance between accessibility and security is paramount. Sharing the common mechanism with external experts for review and constraining the common set of sequences to those that are publicly available are alternative strategies under consideration. 76 As biotechnological tools become more accessible, the urgency of developing safeguarding frameworks and technologies cannot be overstated.
International organizations such as the World Health Organization (WHO), the United Nations (UN), and the UNESCO play pivotal roles in shaping biosecurity norms, ethical guidelines, and global governance frameworks. 77 The WHO can guide member states on biosecurity measures and best practices, 78 and UNESCO (International Bioethics Committee) contributes to the ethical discourse surrounding biotechnology. 79
Collaborative initiatives bring together diverse stakeholders to address complex biosecurity challenges comprehensively. These efforts involve joint research projects, sharing best practices, and developing common standards and guidelines. 80 For example, initiatives such as the joint effort by the Nuclear Threat Initiative, Wellcome Trust, and the World Economic Forum aim to engage leaders and experts in genomics, virology, synthetic biology, bioethics, security, insurance, and scientific publishing. 81 These diverse stakeholders work collectively to identify, assess, and address biosecurity risks associated with advancing biotechnologies.
National risk assessments may overlook existential risks due to their unique attributes, including a global scope, potential for extreme consequences, and emerging nature. 82 These risks are challenging to quantify and often transcend national borders. Risk assessments are typically presented visually in risk landscapes, which may inadvertently downplay the significance of extreme scenarios. 83 For example, the categorization of risks may set low thresholds for the maximum impact category, effectively deprioritizing lower-probability, high-impact risks. This approach might not adequately capture the unique nature of existential risks, which have a low probability but catastrophic consequences. The Sendai Framework, originally designed to address natural hazards, has become a point of reference for discussions on existential risks. 84 It aligns with the UN’s Sustainable Development Goals and underscores a commitment to safeguarding lives, ecosystems, and the global community from the evolving landscape of biological threats. 85 Ultimately, supporting the Sendai Framework in the realm of biosecurity underscores a commitment to safeguarding lives, ecosystems, and the global community from the evolving landscape of biological threats. 86
CONCLUDING REMARKS: FORGING A COMPREHENSIVE BIOSECURITY PARADIGM IN THE AGE OF AI
The existing international frameworks aimed at preventing biological weapons proliferation and biosecurity policies predominantly emphasize the physical regulation of biological agents, biotechnology tools, and materials. 87 However, these frameworks now seem inadequate and outdated as we confront rapidly evolving hybrid threats that exploit emerging technological innovations. To effectively confront this multifaceted challenge, it is imperative that policy makers embark on a collaborative journey with experts spanning various emerging technology domains. 88 This collaborative endeavor should strive to craft regulatory frameworks adept at proactively addressing, identifying, and mitigating the potent threats arising from the convergence of cybersecurity and biotechnology, commonly known as “Cyber-Biosecurity threats.” 89
The last decade has witnessed profound transformations in the realms of biology and biotechnology. These shifts, underpinned by digitization, automation, and their confluence with the cyber domain, have ushered in new vulnerabilities that may inadvertently lead to unexpected consequences and deliberate exploitations. 90 Regrettably, the full extent of these risks often remains underestimated. Tackling these vulnerabilities and safeguarding against them presents a multifaceted challenge. Current cybersecurity solutions primarily target isolated aspects of biosecurity and cybersecurity, often overlooking the intricate issues arising from their convergence. 91
Given the myriad threats, diverse potential targets, and the variances in their potential impacts, ascertaining the applicability and effectiveness of potential solutions remains a complex endeavor. 92 In addition, the absence of a universal model for securing information systems across the bioeconomy implies that solutions, if weak or prematurely implemented, may address one issue while being inapplicable or potentially exploitable in a different context. 93 These vulnerabilities pervade areas such as biomanufacturing, cyber-enabled laboratory equipment, patient-centric systems, as well as the management of “Big Data” generated from “omics” studies, permeating the entire farm-to-table continuum. 94
Addressing the challenges at the intersection of biosecurity and AI on an international scale necessitates a comprehensive approach. One pivotal strategy involves the adoption of a legally binding AI treaty, akin to the Council of Europe’s treaty, which not only upholds human rights, democracy, and the rule of law but also has the potential to include moratoriums on AI technologies that pose threats to human rights, such as facial recognition. 95 Of particular relevance is Article 11 of the treaty draft, which underscores the need to take measures to safeguard public health and the environment in the context of AI application, emphasizing the critical biosecurity/public health nexus. 96 In addition, Article 17 encourages safe innovation while respecting and encouraging innovation. 97 It is essential to acknowledge that while these treaties provide a promising framework for international AI regulation, each nation must individually ratify them within their domestic legal systems, with provisions for opting out of specific obligations as needed. 98
Nonbinding yet influential guidelines, such as the Organization for Economic Co-operation and Development AI principles, offer valuable policy recommendations, particularly in the health care sector. These principles emphasize the importance of transparency, explainability, security, safety, and accountability in AI systems, thereby reinforcing the biosecurity aspects of AI applications. 99 Initiatives such as the Global Partnership on AI, supported by leaders like Prime Minister Justin Trudeau and President Emmanuel Macron, have the potential to foster global collaboration, research sharing, and information exchange on AI, which is pivotal for addressing biosecurity challenges. 100 Although such organizations may not have published substantial work recently, they hold promise for addressing biosecurity concerns on a global scale. 101
In conjunction with international treaties and principles, the classification of AI applications based on risk levels, mirroring the EU AI Act, ensures that high-risk applications, including those with significant biosecurity implications, undergo rigorous scrutiny and regulation, aligning with the imperative to safeguard public health. 102 Furthermore, technical standards set by organizations such as the International Organization for Standardization play a vital role in standardizing risk management, impact assessment, and AI technology development. 103 These standards are indispensable to ensure that AI systems, especially those intersecting with biosecurity, meet globally recognized quality and safety standards.
Lastly, UNESCO’s voluntary AI ethics framework, which concentrates on ethical impact assessments for AI, provides guidance for the ethical and responsible development and deployment of AI systems with biosecurity implications, thereby reinforcing broader biosecurity and public health objectives. 104 In summation, this multifaceted approach, encompassing legal treaties, international principles, collaborative initiatives, risk-based classifications, industry standards, and ethical frameworks, forges a robust foundation for addressing the intricate biosecurity challenges posed by AI on the global stage.