Back to home

Post

The neglect of agricultural AI-biosecurity: Implications for inter-state conflict

June 2026 / AI biosecurity / agriculture

Introduction

It has become quotidian to describe the upending of another field by artificial intelligence. But in the case of biosecurity, the threat dynamics have fundamentally changed. Artificial intelligence poses a dual threat: large language models such as ChatGPT are capable of uplifting bad actors1, while biological design tools can develop entirely new pathogens5. Focus in this area has been especially on human pathogens, unsurprising given the recent COVID-19 pandemic110. Human pathogens are difficult to weaponise, as a smallpox epidemic is likely to eventually infect the population of the state that deployed it8. Agricultural pathogens, however, have different dynamics, which may make them more appealing as targets for bioweapons research11. This, together with AI developments, represents a greater threat potential.

Biosecurity and artificial intelligence

Biosecurity involves protecting the population and the food we eat from pathogens9. The two sides of this defence are protecting against pathogen spillover, such as the 2019 coronavirus, and protecting against bad actors, such as the Japanese doomsday cult Aum Shinrikyo2. Developments in AI large language models (LLMs) such as ChatGPT or Claude have lowered the bar for developing a bioweapon2, thereby exacerbating risks associated with bad actor-driven bioweapon development.

Many preventive countermeasures have focused on preventing bad actors, such as terrorists, by making it more difficult to access other components needed for bioweapons development. State actors have been theorised to be unaffected by LLMs, since they have no issue with accessing funding and technical know-how for developing bioweapons3. Correspondingly, bioweapons research is mostly led by states3.

However, there is another side to AI-enabled bioweapons development: AI-enabled biological design. Biological design tools are AI models trained on biological data and have enabled complex molecular design. It is now possible to design proteins never found in nature with specific desired properties5. However, these capabilities have also opened the door to de novo pathogen design4. The concern is that this constitutes a capability previously unpossessed by states. De novo pathogen design, therefore, represents a real increase in the ability of states to develop bioweapons.

Agricultural biosecurity in the age of AI

A key reason biological warfare is so rarely used is the potential for the boomerang effect, where pandemics infect the population of the deploying state, but this is not a strong feature of agricultural pathogens. The possibility of this happening is significantly reduced among agricultural pathogens. Agricultural pathogens are plant-specific, infecting only specific species, such as wheat or corn7. Additionally, cereal crop make-up is very different between different states; Russia and Europe are dominated by wheat and rye, while China primarily produces rice and America is the world leader in corn6.

Coupled with lower transmissibility among plant pathogens, this makes plant pathogens an ideal target for state bioweapon programs. A rice-specific bioweapon, for instance, will only target the countries that grow a large amount of rice7. Transmission between states is lower, as plants are necessarily sessile, minimising inter-state or inter-continental transmission, at least compared to viruses such as COVID-19. Even with high transmissibility, corn, wheat, and soybeans are fundamentally different plants, with pathogens having limited ability to infect hosts they are unadapted to. The strategic appeal is not simply that crop pathogens may be host-specific, but that their effects could plausibly appear as natural outbreaks while imposing economic, food-security, and political costs on an adversary.

Conclusion

AI's biosecurity risks are not uniform. LLMs may lower barriers for sub-state actors, but biological design tools give states a more significant new capability: the ability to design novel pathogens. This matters most in agriculture, where the boomerang effect that limits human bioweapons is weaker. Because agricultural pathogens can be crop-specific and geographically targeted, they may be more strategically attractive to states. Treating agricultural biosecurity as secondary therefore risks overlooking one of the most plausible ways AI could reshape interstate biological conflict.

References

  1. Götting et al. (2025) Virology Capabilities Test (VCT): A Multimodal Virology Q&A Benchmark. https://arxiv.org/abs/2504.16137
  2. Rosenau, W. (2001) Aum Shinrikyo's Biological Weapons Program: Why Did it Fail? Studies in Conflict & Terrorism, 24(4), 289-301. DOI: 10.1080/10576100120887
  3. Guillemin J. (2006) Scientists and the history of biological weapons. A brief historical overview of the development of biological weapons in the twentieth century. EMBO Rep. Jul; 7 Spec No: S45-9. DOI: 10.1038/sj.embor.7400689
  4. King, S. H. et al. (2025) Generative design of novel bacteriophages with genome language models. bioRxiv 2025.09.12.675911. DOI: 10.1101/2025.09.12.675911
  5. Thomas Hayes et al. (2025) Simulating 500 million years of evolution with a language model. Science 387, 850-858. DOI: 10.1126/science.ads0018
  6. FAO. (2024) Agricultural production statistics 2010-2023. FAOSTAT Analytical Briefs, No. 96. Rome.
  7. Morris & Moury (2019) Revisiting the Concept of Host Range of Plant Pathogens. Annu Rev Phytopathol. 25;57:63-90. DOI: 10.1146/annurev-phyto-082718-100034.
  8. Casadevall & Pirofski (2004) The weapon potential of a microbe. Trends in Microbiology, 12;259-263. DOI: 10.1016/j.tim.2004.04.007
  9. Meyerson & Reaser (2002) Biosecurity: Moving toward a Comprehensive Approach. BioScience, 52:7; 593-600. DOI: 10.1641/0006-3568(2002)052[0593:BMTACA]2.0.CO;2
  10. Webster et al. (2025) Global Risk Index for AI-Enabled Biological Tools. The Centre for Long-Term Resilience & RAND Europe. https://doi.org/10.71172/wjyw-6dyc
  11. Feldman, Jonathan and Feldman, (2026), Agricultural Biosecurity in the Age of Biological AI. SSRN. DOI: 10.2139/ssrn.6446998