The threat of biological weapons has received renewed attention following the publication of Anthropic’s threat report on 10 September 2026, which provided five case studies on biological misuse. These were the first publicly documented examples of attempted biological misuse of AI models covered in detail, including the safeguards in place to disrupt such efforts.
The broad diffusion of AI across scientific disciplines is increasingly evident in bibliometric studies. The number of AI-related natural science publications rose to nearly 80,000 in 2025, up 26% from 2024. This accounted for roughly 6–9% of scientific output across major scientific domains in 2025, compared to less than 1% in 2010. Recent studies also highlight particularly rapid growth in AI-biotechnology research, including in areas such as gene editing, drug discovery and bioinformatics.
These trends suggest that once-niche AI models are becoming increasingly embedded across the scientific research ecosystem. No longer the preserve of specific communities in specialized domains, these models are evolving into general-purpose capabilities available to specialists and non-specialists alike. This expanded access and rapid improvement in capability suggest significant implications for the governance of dual-use biology and its threat outlook.
Biological threats, including biological weapons, have traditionally been associated with access to specialized material, facilities and expertise. The growing integration of AI models into biological research and their rapid advancement add another dimension to these threat perceptions. Frontier AI models are increasingly enabling actors to acquire and apply biological knowledge more efficiently than was possible just a few years ago, potentially lowering some of the knowledge and expertise barriers associated with sensitive dual-use research.
Though Anthropic’s findings provide evidence that actors have already attempted to misuse AI for sensitive and potentially weaponized biological research, they do not demonstrate that AI models enable the development of biological weapons, nor that such efforts would have meaningfully advanced the prospect of success. They do, however, suggest that actors see strategic value in exploiting frontier AI models for dual-use biological activities that may be inaccessible through traditional pathways.
For that reason, it is encouraging to see frontier AI companies engaging in multilateral discussions on threat models, biological risk assessment and evaluations with respect to their tools. This is particularly important given that the private sector has been responsible for the development of the overwhelming majority of frontier AI models. But key governance and oversight challenges remain, and the most capable models continue to exhibit the least transparency.
At the same time, the international community cannot become wholly reliant on private sector actors to identify and mitigate every single situation in which AI could be exploited in the development of biology for hostile purposes. Nor can the private AI industry afford to remain detached from ongoing multilateral efforts to strengthen biological weapons controls.
Effective governance of emerging AI-biology dual-use capabilities will require sustained coordination with existing biological weapons prohibition and biosecurity frameworks, including the core, near-universal regime governing biological threats, the Biological Weapons Convention (BWC). Efforts to strengthen the BWC’s ability to address emerging science and technology developments may ultimately prove to be relevant to the broader debate on AI governance, both within industry and among governments.
Moreover, the convergence of AI and biology, while deserving particular attention, is only one of several converging science and technology areas that may shape the biological threat spectrum and warrant our attention as we look to avert a darker biological weapons future. In such a future, biology could be exploited to develop threats beyond traditional conceptions of biological weapons, including localized attacks, terrorism and economic sabotage via disruption of agriculture or livestock supplies.
The nature of biological enablement by AI raises both the floor and ceiling of what could be achieved. Novices can attain previously inaccessible scientific knowledge, whereas experts can use AI to overcome domain-specific hurdles in their research. And a wide range of actors could exploit AI to facilitate mis- and disinformation related to biological weapons.
Averting such a future will require approaches that address the broader convergence of emerging science and technology developments rather than focusing solely on any single capability or technological phenomenon. More importantly, no one safeguard would be sufficient. A layered approach is needed to address the rapidly evolving scientific landscape in the context of varying real-world circumstances.
A web of prevention
Fortunately, States have developed several measures to prohibit and prevent biological weapons. Particularly important at the international level is the 1972 Biological Weapons Convention. This agreement prohibits, among other things, the development, production and stockpiling of biological weapons, which are defined broadly under the convention to include any form of biological weapons, including any biological weapon developed using AI. With 189 States Parties, the treaty covers the vast majority of the world. The BWC has served States well for more than half a century; no State admits to having a biological weapons programme, and allegations of association with biological weapons have been met with sharp denials.
Moreover, many States have transposed this international prohibition into national law, as well as a range of other regulatory and normative measures designed to prevent the re-emergence of biological weapons. This process of national implementation via measures that prohibit through national laws and prevent through wider technological governance mechanisms is particularly important. It helps to construct both a “web of prevention” and communities of compliance that can together bolster the prohibition on biological weapons.
International and national implementation of measures to prevent biological weapons needs to be nurtured to ensure that these measures remain fit for purpose in the age of digital biology. Indeed, Anthropic’s threat report provides details of cases in which users took evasive measures to hide their tracks and circumvent existing safety mechanisms. Implementation strategies must therefore be developed with varied obfuscation methods in mind.
A decisive moment for the Biological Weapons Convention
Despite a number of concrete steps undertaken nationally and internationally, the BWC continues to have several operational limitations. It has no verification mechanism, for example, even though States tried throughout the 1990s to negotiate a protocol with a verification component. Moreover, in contrast to treaties on chemical and nuclear weapons, which are supported by organizations employing hundreds or even thousands of staff, the BWC is supported by a four-person Implementation Support Unit hosted within the United Nations Office for Disarmament Affairs.
The Working Group on strengthening the BWC will meet for its tenth and final session in December 2026. Over the course of this process, a draft report has emerged that contains a package of measures designed to advance the treaty. These include a plan for further work on verification through a dedicated open-ended working group, a mechanism for reviewing science and technology, and a second mechanism focused on international cooperation and assistance. These measures have considerable potential to advance the BWC and strengthen biological security.
Success in the Working Group would pave the way for the upcoming BWC Review Conference in 2027, during which States and stakeholders will have the opportunity to chart a course towards effective prevention of biological weapons for another 50 years.
But if the Working Group fails to achieve consensus in December, there is a real risk that biological disarmament diplomacy could be set back 25 years, to the doldrums of 2001, when negotiations on a protocol to strengthen the BWC collapsed acrimoniously.
A similar collapse of the Working Group may not irreparably damage the BWC or the current prohibitions on biological weapons, but it would severely curtail momentum towards strengthening the BWC. Avoiding such a fragmentation of multilateral efforts by States and stakeholders would help us tackle one of the most pressing security challenges of our time.
