What this threat is
Biological weapons use living organisms, or the toxic substances produced by them, to harm people, animals, or plants. They're prohibited under the 1972 Biological Weapons Convention, which virtually all countries have signed. But treaties can only do so much when the underlying science becomes widely accessible. The concern with AI is not that it creates an entirely new threat, but that it dramatically accelerates the diffusion of knowledge and capability that makes biological weapons easier to develop.
Modern AI tools, particularly large language models trained on scientific literature and AI systems designed for drug discovery and protein design, can compress the research and development process that once required teams of expert scientists. They can suggest experimental approaches, identify which genetic modifications might increase a pathogen's transmissibility or lethality, and help troubleshoot laboratory techniques. In 2022, researchers published a study showing that an AI drug-discovery tool could, when its safety constraints were reversed, suggest thousands of potentially toxic molecules, including chemical weapon agents, in under six hours. The same dynamic applies to biology.
The threat has multiple dimensions. A state-level actor could use AI to design pathogens with capabilities beyond what natural evolution has produced. A sub-state group with access to desktop DNA synthesis equipment and AI tools might be able to recreate or modify dangerous viruses. A lone individual with the right knowledge and equipment, knowledge that AI might help them acquire, represents a category of risk that biosecurity frameworks were not designed to address.
What makes this particularly difficult is the dual-use nature of the underlying science. The same tools that help design pathogens can help design vaccines. The same AI systems that could be misused to engineer a dangerous virus are being legitimately used to develop treatments for cancer and genetic disease. You can't simply prohibit the tools without prohibiting beneficial research too.
Why it matters
Unlike most weapons, biological agents can self-replicate and spread. A successful attack with an engineered pathogen could seed a pandemic, with consequences that scale far beyond the initial release. The historical record of natural pandemics illustrates the scale of harm a highly transmissible pathogen can cause. An engineered pathogen, optimized for transmissibility or lethality, could be worse.
The other dimension is irreversibility. Once an engineered pathogen is released into the world, you can't recall it. Early detection and rapid response are critical, but both depend on systems that need to be built before an attack, not after. Current biosurveillance infrastructure in most of the world is not adequate for the speed at which AI-enabled bioweapon development could move.
Where things stand today
The AI biosecurity risk has received significant attention from researchers and some governments. In the US, the Executive Order on AI specifically called out biological risks and directed agencies to assess them. Several leading AI labs have developed policies around withholding or restricting models from providing certain types of biosecurity-relevant information. The Biological Weapons Convention has modernization discussions underway, though treaty enforcement remains difficult.
The scientific and biosecurity communities have proposed several approaches: screening DNA synthesis orders to prevent the production of dangerous sequences, developing better biosurveillance systems that can detect unusual pathogen activity early, and creating governance frameworks for dual-use life sciences research. Progress on all three fronts exists but remains insufficient relative to the pace of AI capability development.
How Better Societies helps
Summit: Biosecurity and AI governance are natural partners, and the Better Societies Summit brings these communities together. Understanding the intersection of AI capability and biological risk is essential for anyone working on either problem.
Compliance: Organizations developing or deploying AI systems with potential life sciences applications face specific obligations under the EU AI Act. Our Compliance advisory helps organizations understand where their systems fall in the risk taxonomy and what documentation and safeguards are required.
Accelerator: The Accelerator supports founders working on biosurveillance, pandemic preparedness, and biosecurity tools. This is one of the highest-leverage areas for AI safety work.