In 2025, Stanford University PhD student Samuel King accomplished what biosecurity experts have long feared: he used a generative AI model to propose functional genetic blueprints for microscopic viruses. The experiment wasn't designed to create actual pathogens—King stopped short of synthesis—but rather to demonstrate that large language models trained on publicly available genetic databases could generate novel viral sequences without meaningful friction or oversight. The sequences King's model produced were not theoretical abstractions but plausible genetic instructions that, if synthesized, could potentially produce infectious agents. This capability emerged not from specialized bioweapon development systems but from standard commercial AI tools, raising an uncomfortable question: if a Stanford researcher can do this in an afternoon, what prevents a bad actor with similar skills from doing the same?

The regulatory architecture supposed to govern dual-use biological research—technology that can serve legitimate or harmful purposes—was constructed for a slower, more gatekept world. The National Institutes of Health maintains institutional biosafety committees that review research proposals, while the State Department enforces International Traffic in Arms Regulations (ITAR) restricting export of sensitive biotechnology. Yet these frameworks assume human researchers at established institutions conducting defined experiments. They have no mechanisms to monitor or restrict AI models' outputs, which are generated invisibly and at scale. A researcher querying an AI chatbot leaves no institutional footprint. The FDA, NIH, and Department of Defense lack clear authorities to preemptively regulate which genetic sequences AI systems can be trained on or prompted to generate. Industry self-regulation through responsible disclosure remains fragmented, with no mandatory reporting requirements when models produce concerning outputs.

The stakes are escalating rapidly. Generative AI accessibility has exploded exponentially over the past two years; what required specialized knowledge in 2023 now requires only prompt engineering skills. Unless concrete policy changes materialize within the next 18 months—including mandatory biosecurity reviews before deploying large biological datasets into public AI models, real-time monitoring of AI-generated genetic sequences by intelligence agencies, and international coordination on synthetic biology export controls—the window for preventing widespread access to pathogenic design capabilities will effectively close. The question facing policymakers is no longer theoretical: Do we regulate AI-generated biology now, or do we wait until an attack forces retroactive response?