Public discourse often centers on artificial intelligence as an existential hazard, yet a growing body of analysis highlights the immediate danger of engineered pathogens. Earlier this month Anthropic released a misuse report that catalogued five instances in which foreign researchers attempted to skirt the model’s safeguards to obtain gain-of-function information. The document illustrates how AI tools can be repurposed to aid bioweapon design.
Gain-of-function work involves altering a microbe’s genome to increase its infectivity, lethality or host range. The article’s author, trained as a molecular biologist, notes that scientists studying high-risk agents frequently argue that publishing their findings outweighs the risk of accidental or intentional release. This mindset persists despite the potential for catastrophic outcomes.
Annie Jacobsen’s recent volume, *Biological War: A Scenario*, draws on her long-standing contacts within the defense establishment to map the threat landscape. A declassified Defense Department assessment places nuclear weapons at the extreme end of the mass-destruction curve, but ranks biological weapons as a close second because the barriers to entry are minimal. Jacobsen argues that this low threshold makes bioweapon development more probable than nuclear proliferation.
Unlike nuclear arms, which require fissile material and delivery platforms such as intercontinental ballistic missiles, bombers or submarines, biological agents can be produced in laboratory settings. Officials at the CIA and other intelligence agencies maintain active counter-proliferation programs aimed at disrupting illicit bioweapon projects. Nonetheless, experts contend that accidental releases from secure facilities present a far greater probability of a global health crisis than a deliberate terrorist strike.
The world now hosts roughly 3,600 biosafety level 3 and 4 laboratories, facilities designed for work with the most dangerous pathogens. By contrast, the number of terrorist groups capable of fielding a bioweapon is far smaller. Historical cases illustrate the accident risk: a Soviet scientist inadvertently inoculated himself with Marburg virus and succumbed to hemorrhagic fever, while other researchers have suffered similar mishaps after self-injection or equipment failure.
Advances in genetic engineering and AI-driven design have expanded the toolbox for creating highly transmissible agents. Experiments such as the mousepox study that rendered a virus resistant to existing immunity, and the H5N1 ferret-flu work that produced airborne transmission, demonstrate how modest modifications can dramatically alter pandemic potential. Some investigators have expressed confidence in publishing full genomic sequences of lethal constructs, a stance critics label as dangerously hubristic.
The Anthropic misuse report cited earlier documents foreign scientists seeking to exploit the Claude language model for gain-of-function queries, underscoring how AI can lower technical barriers. The five recorded attempts show that even modest AI assistance can provide the detailed guidance needed to manipulate pathogen genomes, raising alarms among biosecurity officials about the ease of illicit research.
Jacobsen’s book deliberately omits discussion of AI-enabled biology, but interviewees warned that the technology further erodes the already low entry threshold for bioweapon creation. The convergence of high-containment labs, gain-of-function research, and readily available AI tools creates a scenario where accidental release or deliberate misuse could trigger a global health emergency.