Biosecurity in the Age of AI

The Human Genome Project revolutionized our understanding of DNA and ushered in a new era in genetic exploration. Similarly, the fields of gene and recombinant protein synthesis – the processes by which designed sequences are transformed into physical instantiations – have undergone groundbreaking changes leading to increasingly lowered barriers to entry in synthetic biology. (Like sequencing, gene synthesis is orders of magnitude more affordable than ever before.)

These developments will impact a diverse array of sectors, with notable benefits for virus identification and engineering. However, the democratization of biological tools and knowledge, propelled by the integration of AI tools like LLMs, will also empower a growing number of actors with the capability to create or enhance pandemic-scale biological threats. Consider the following example:

In a recent academic exercise, an AI chatbot assisted graduate students in identifying four potential pandemic pathogens, issued step-by-step “lay-person” instructions for virus generation, and directed students toward companies that could manufacture synthetic DNA sequences for the purposes of engineering and producing these pathogens. The entire exercise took an hour to complete. While such a set of instructions could not enable an untrained actor to engineer a pandemic-capable pathogen today – and it is possible that with additional training and time, the same results may have been achieved without AI – the example showcases the potential of technology to “upskill” individuals without relevant subject matter expertise or experience. By leveraging a small slice of well-intended research, ongoing technological advancement will also increase the overall “ceiling” of harm by rendering pathogens more transmissible, deadlier, and/or more able to evade existing vaccines or therapies than their natural counterparts.

AI-enabled biology has and will continue to furnish world-changing breakthroughs for human health and disease management, offering opportunities in pandemic preparedness, cancer detection and treatment, and chronic disease management. It also poses significant challenges. As innovation progresses and industry accommodates demand at the AI Bioconvergence nexus, Helena is working to evaluate biosecurity risks, recommend safeguards, and advocate for enhanced, collaborative efforts to mitigate against emerging threats.

In January of 2023, Helena undertook a 6-month biosecurity risk landscape analysis, consulting with nearly 100 experts across academia, government, technology, public health, and the private sector. A significant subset oriented us to heightened risks related to the sequencing, characterization, design, engineering, and manufacturing of genetic material and pathogens, and pointed to limitations in oversight creating biosafety risks (relating to accidental release), biosecurity risks (relating to intentional misuse), and information risks (relating to the publication or other communication of materials that could be used to cause harm).

Recognizing the exponential growth of emerging technologies and the imperative to urgently address novel threats, Helena convened a group of senior leaders from industry, government, think tanks, and academia to interrogate this risk landscape and pressure-test courses of action. Their conversations took place over the course of two and a half days at The Rockefeller Foundation’s Bellagio Center.

The meeting was informed by extensive review of existing policy frameworks and subject matter literature. While primarily focused on emerging threats related to AI, the group also addressed risks relating to Dual Use Research of Concern (DURC) and research with enhanced Pathogens of Pandemic Potential (ePPPs).

These discussions resulted in the Biosecurity in the Age of AI Chairperson’s Statement, authored by the Honorable Mark Dybul, MD, Helena Member, with robust input from the Bellagio attendees.

The report recognized the complexity and pace of developments in AI-enabled biology and emphasized critical areas requiring further reflection, ongoing monitoring, and long-term analysis.

To advance critical action in the near-term, the Biosecurity in the Age of AI Chairperson’s Statement also recommended the following key interventions:

• The immediate establishment of Public-Private AI Task Forces and subordinate, technical working groups focused on specific AI risks at the highest executive levels across countries, regions, and multinational entities. To continuously monitor and respond to evolving threats and future-proof the emerging technical landscape, Task Forces and Working Groups should integrate cross-sectoral expertise, including advanced biosecurity expertise, and develop mechanisms for regular collaboration across agencies, governments, and international entities. To defend against undue influence and incentive misalignment, Task Forces must be protected through both funding sources and operating structure.

• Safeguarding the digital to physical frontier. Governments, especially those with advanced synthetic biology and AI economies, should work closely with the private sector to prevent digitally-designed threats from transforming into physical threats. AI technologies can enable discovery of harmful biological functions and furnish pathways for developing them, including by actors who may do so unsafely or with harmful intent. Therefore, the digital-to-physical boundary is increasingly fragile yet it remains the primary safeguard against misuse. As an essential first step, governments should implement mandatory screening policies for DNA synthesis. Given advancements in “desktop synthesis” that create critical vulnerabilities in the digital-to–physical frontier, Technical Working Groups should concertedly assess and develop screening tools and policies that extend beyond publicly funded and industry-level actors and address emerging threats from individuals operating benchtop synthesizers and assemblers. To safeguard the digital-to-physical frontier over the long-term, governments should integrate blue-. yellow-, and red-teaming approaches to surface and remediate vulnerabilities in screening mechanisms and invest in “next generation” tools and methodologies that counter efforts to bypass standard screening approaches.

• Appropriately Guardrailing LLMs, BDTs, and emerging technologies. The use of AI tools in synthetic biology will expand access to pandemic-class biological agents and will allow a growing number of actors to enhance the lethality, host range, or transmissibility of these pathogens. Therefore, governments, biotechnology developers, and life science research funders should develop approaches to rapidly guardrail these technologies. Task Forces should consider whether and how to control access to powerful AI models – perhaps through Know Your Customer and Know Your Order policies linked to operational entities with enforcement capabilities. Technical Working Groups should also develop approaches to test and evaluate AI models and thoughtfully consider accountability mechanisms, as well as potential liability and regulatory measures, that encourage the private sector to responsibly develop and deploy technologies. As with screening mechanisms, Working Groups should implement blue-, yellow-, and red-teaming approaches to monitor emerging threats and evolve guardrails over time as technologies advance. In concert with Technical Working Groups focused on information integrity, AI Bioconvergence Working Groups should also address threats posed by AI to fuel mis- and disinformation in biology, which could undermine confidence in the economic and functional value of biotechnologies and stoke chaos in the event of a biological incident.

• Refining ePPP policies and reinforcing biorisk policies to protect against accidental and intentional misuse. Governments should update definitions of concerning pathogens to accommodate technological advances in gene synthesis and manufacturing, increase oversight of research with enhanced Pathogens of Pandemic Potential, and establish independent review mechanisms to enable more effective risk reduction in the field. Health authorities and their advisory bodies, as well as appropriate academic bodies, should advance additional efforts to improve, extend, and evolve oversight policies to address AI Bioconvergence. In addition, governments should engage with leading biosafety and biosecurity experts to decrease risks of accidental infection/transmission and deliberate misuse and ensure ongoing oversight and surveillance of novel research, biotechnology development, and science. Tracking measures should detect, assess, and prevent immediate threats, while also monitoring unforeseen effects over the long-term. Enhanced oversight measures should include increasing transparency around biosafety and biosecurity protocols, resourcing entities performing biological research and disease surveillance to build sufficient biosafety and biosecurity capacity, establishing graduated reviews to triage research according to risk level, and conducting risk/benefit analyses to identify and fund alternatives to highest-risk research.

• Enhancing biosecurity and biosafety norms to include AI-Enabled Biology and promote International organizations and tools to practically implement them. The swift pace of technological development necessitates the evolution of biosafety and biosecurity norms, standards, and practical implementation. For decades, national and international tools to reduce biological risks have lagged significantly behind technology development. AI Bioconvergence advances are the latest to surge past existing risk reduction frameworks. To meet and get ahead of emerging risks and opportunities, biotechnology and life sciences research funders must prioritize biosafety, biosecurity, and AI Bioconvergence as an integral component of their mandate. In addition, Technical Working Groups should recommend rigorous risk/benefit assessments in review processes for AI Bioconvergence research that carries the potential to cause large-scale harm. In concert with governments and key advisors, TWG’s should also seek to develop norms to address new tensions surfaced by AI Bioconvergence, such as the risks inherent to the development and dissemination of data sets. New norms should be developed and pressure tested in accordance with best practices from cybersecurity and nuclear security where relevant. In tandem, research and technology funders should commit to regular biosafety and biosecurity reviews and build additional funding into proposal and investment costs to support biosecurity-by-design approaches and accommodate more robust safety measures and requirements. The creation of innovative tools that allow stakeholders from across government, foundations, and the private sector to mitigate risk in real time – while new technologies are being developed – will be essential.

• Invest in early warning and detection, response capacity, and accountability measures, and build biosafety and biosecurity into these approaches. Many experts have warned about the possibility of another naturally-caused pandemic at the magnitude of COVID-19 within the next decade. Given added risks from AI Bioconvergence – including accidental and intentional release of synthetically-created pathogens – it is therefore imperative that governments, regional, and multinational entities strengthen surveillance and response capacity worldwide. Governments and entities across continents have taken steps to bolster health security regulations, enhance early warning and detection systems, invest in PPE and medical countermeasures, and set aside financing to ensure readiness in the event of a pandemic.

The full report can be accessed below.

Though many of those in attendance contributed their thoughts under Chatham House rule due to their positions within government, The Biosecurity in the Age of AI Report was publicly endorsed by a group of leading experts. A full list of endorsees can be found here.

The Biosecurity in the Age of AI Chairperson’s statement and related discussions have influenced U.S. executive orders on AI, Senate testimony on AI risks, and reports released in response to the UK AI Safety Summit. The team has been asked to brief a range of policymakers and industry leaders, including senior members of the State Department, the Department of Homeland Security and experts such as Microsoft AI CEO Mustafa Suleyman.

Significantly reducing biosafety and biosecurity risks without meaningfully curtailing scientific progress will require a combination of immediate, concrete actions alongside long-term analysis. Helena continues to focus on governance, legislative advocacy, and creating fora for meaningful dialogue to help develop adequate, complete, and holistic approaches for a safe and resilient high-tech future.