The convergence of artificial intelligence and modern biotechnology is forcing a long-deferred reckoning with the weakest link in the weapons-of-mass-destruction control architecture. Biological weapons have been banned since the 1975 Biological Weapons Convention, ratified by 189 countries — but unlike the chemical weapons regime enforced by the OPCW with inspections and destruction oversight, the bioweapons ban operates on transparency, voluntary reporting, and good faith. There is no inspectorate, no verification mechanism, and no enforcement body. The warning signals are sharpening. In October 2026, MI5 chief Ken McCallum stated publicly that AI could increase the risk from existing or new biological weapons, including from non-state actors. Separately, Anthropic reported in September 2026 that it had identified and blocked multiple attempts to use its AI assistant Claude for possible biological weapon research — including grant applications for gain-of-function research and projects involving avian influenza variants with high mammalian transmissibility. Anthropic itself acknowledged the dual-use problem: the same research could serve legitimate medical prevention. The dual-use dilemma is the structural core of why biological weapons are so difficult to control. The laboratory equipment used to develop vaccines and medicines is indistinguishable from equipment that could support military research. A biological attack can initially present as a natural disease outbreak, and only extensive forensic investigation can distinguish deliberate release from natural emergence. This makes detection fundamentally harder than for nuclear or chemical programs, where production signatures are more distinctive. Historically, the threat has been real but constrained by technical barriers to weaponization. AI changes the calculus by dramatically accelerating the analysis and manipulation of pathogen genomes. What previously required teams of specialized researchers with years of training can now be partially automated — lowering the skill floor for potential bad actors while simultaneously advancing legitimate medical science. The speed at which AI can evaluate biological data means that the window between a novel pathogen being sequenced and someone understanding how to make it more dangerous is compressing. The governance gap is stark when compared to peer regimes. The Chemical Weapons Convention created the OPCW, which conducts inspections and oversees stockpile destruction. The Nuclear Non-Proliferation Treaty is backed by the IAEA's safeguards system. The Biological Weapons Convention has neither. States have resisted inspection mechanisms because they fear exposing sensitive research data from their medical, biotech, and pharmaceutical industries — a reasonable concern that has produced an unreasonable outcome: a ban with no teeth. The 2001 US anthrax letters — which infected 22 people and killed five just weeks after September 11 — demonstrated that even a single actor with access to laboratory-grade material could weaponize a biological agent within an advanced nation's borders. That incident used a pathogen well understood since the mid-20th century. The question now is what happens when AI-assisted biotechnology makes it possible to engineer pathogens with novel characteristics — enhanced transmissibility, vaccine resistance, delayed onset — at a pace that outstrips the capacity of existing public health and intelligence systems to respond. The international community faces a familiar but accelerating problem: the technology is moving faster than the governance. The BWC's next review conference will determine whether states can overcome their resistance to verification mechanisms before AI-enabled biotech capabilities diffuse beyond the point where treaty architecture alone can contain the risk. The 50-year gap between the convention's entry into force and the emergence of AI-accelerated bioengineering may prove to be the most consequential period of institutional neglect in modern arms control.
