Can AI Protect Us from Bioweapons? by Dr. Timothy Smith

Photo Source: GetArchive

Recently, on September 23, 2025, President Trump, in an address to the United Nations, announced that the US government will develop artificial intelligence to help protect us from bio-weapons. Biological weapons include the gain-of-function modification of known pathogens, such as the development of more virulent forms of existing diseases like influenza, as well as the creation of novel pathogens engineered for lethality. The mention of AI as a tool for security against bio-terrorism development and attacks from hostile countries made me think about how and if AI can help improve our security in a world where the barrier to building bioweapons is lowering with more advanced AI models becoming available to the public on an almost daily basis.

Traditionally, protection from bioweapons has come from the deep expertise required across multiple disciplines, including microbiology, virology, DNA synthesis, fermentation, and synthetic biology, necessary to develop and distribute a bioweapon. The compound expertise required to make a weapon made it extremely difficult for rogue actors to assemble, fund, and equip such a team with the proper laboratory equipment and safety apparatus. However, such an endeavor can happen, as evidenced by the Japanese doomsday cult called Aum Shinrikyo that killed 14 people and injured thousands of others in a Tokyo subway attack with sarin gas in 1995. The terrorists manufactured the sarin in a secret chemical weapons facility for use in their attack with the help of a chemist named Masami Tsuchiya. (forensicscience.com)

Today, AI models such as ChatGPT and biology-focused models like Evo-2 significantly reduce the expertise required to develop biological weapons, offering step-by-step instructions and potential assistance in weapon design. Evo-2, developed by Arc Institute, a nonprofit biomedical research organization, and NVIDIA, with contributions from Stanford University, UC Berkeley, and UC San Francisco researchers, features an LLM designed to help researchers develop new therapeutics from predicting the effects of mutations in various animals, including humans, to generating new genetic sequences for developing synthetic organisms. The purpose of synthetic organisms varies from creating new bacteria designed to produce valuable drugs to bacteria that can digest pollutants or break down microplastics into safer pieces.

The noble goals of Evo-2 also make it a potential tool for evil in the hands of a malevolent actor, who could design biological weapons. Although the designers of Evo-2 removed specific pathogens that affect humans and other higher animals from the training data, Bad actors could retune the model with pathogenic information that would allow the model to answer questions such as what mutations in an influenza virus would make it more virulent. Before the release of tools like Evo-2, only deep specialists could provide a good answer to this question, but now anyone can get a satisfactory response. Such information, in the wrong hands, along with a powerful chatbot, could provide the methods for an average scientist to develop dangerous weapons.

Given the diminishing skills needed for terrorists or rogue states to develop bioweapons, security specialists need help from AI to monitor and predict who would develop such weapons and where they might be. AI can help in several ways, including supply chain oversight, integrating open-source data analysis, and advanced biosurveillance. AI can sift through billions of data points to find patterns that indicate bioweapon development—for example, by analyzing supply chain events through the tracking of specialized lab equipment sales, such as DNA synthesizers, and comparing them against export control lists to identify anomalies. Flagging equipment going to unusual locations could help prevent sensitive equipment from falling into the wrong hands. Additionally, tracking genetic sequence orders for potentially dangerous sequences by cross-referencing against known hazardous sequences can also be beneficial. Tracking open-source model usage for suspicious queries. For example, AI developers like OpenAI already implement "always-on detection systems" to block or flag suspicious queries, particularly those concerning biological synthesis or pathways for weaponization. 

AI-powered biosurveillance systems could analyze global health and climate data and monitor for epidemiological anomalies. AI could sift through clinical data, wastewater monitoring results, and genetic sequencing information from pathogens to detect the early emergence and spread of engineered pathogens.

With the help of AI, our security groups can monitor and predict potential threats from bioweapons. However, in many ways, a clumsy terrorist or rogue state would likely reveal their intentions by using publicly available systems, models, and services. However, the more careful bad actor would know to disguise their activities through legitimate labs and underground facilities, and fine-tune their models offline. As with many advances, new technologies such as AI may accelerate our ability to find new therapeutics for devastating diseases, but with these new advances, bad actors can also turn the same tools against humanity for evil purposes. Vigilance, not only at the AI model level but also at the societal level, must be on the mind of every citizen to prevent the development of bioweapons.

Dr. Smith’s career in scientific and information research spans the areas of bioinformatics, artificial intelligence, toxicology, and chemistry. He has published a number of peer-reviewed scientific papers. He has worked over the past seventeen years developing advanced analytics, machine learning, and knowledge management tools to enable research and support high-level decision making. Tim completed his Ph.D. in Toxicology at Cornell University and a Bachelor of Science in chemistry from the University of Washington.

You can buy his book on Amazon in paperback and in kindle format here.