AI-enabled bioweapons: The looming threat that biotech can’t ignore

The rapid rise of AI-enabled bioweapons is prompting a rare chorus of alarm among the sector’s most senior figures.

Why AI-enabled bioweapons demand urgent attention

In late March, Anthropic’s chief executive Dario Amodei warned that unchecked progress could unleash existential danger, a sentiment echoed by OpenAI’s Sam Altman on X. “We must pace the frontier,” Altman wrote, underscoring a shared urgency. A week earlier, former Anthropic researcher Jacob Coxon quit, accusing both Anthropic and OpenAI of reckless complacency, while colleague Evan Hubinger tweeted that the probability of AI killing humanity exceeds ↓ 10% within a decade. The most chilling scenario involves AI accelerating the design of lethal pathogens – from gene‑targeted viruses to toxin‑laden fungi – that could be weaponized with unprecedented speed. In 2022, a team at Collaborations Pharmaceuticals demonstrated that an AI molecule generator could produce ↓ 40,000 candidate compounds for chemical warfare in under six hours, some surpassing known nerve agents in toxicity.

“That was very scary to me,”

recalls Stanford bioethicist David Magnus, who has warned for years about biotech misuse. Today’s large‑language models can field detailed protocols on synthetic biology, effectively handing the knowledge of “almost every scientist who ever lived” to anyone with internet access, notes biosecurity scholar Dunja Sabra of the University of Hamburg. When paired with the DIY‑biology movement, which enables home‑grown labs, the risk of a determined actor succeeding becomes more than theoretical. Existing safeguards – DNA‑order screening, red‑team/blue‑team reviews, and AI providers’ content filters – are far from foolproof. Anthropic recently disclosed attempts to misuse its models for enhancing chikungunya transmissibility, engineering deadlier avian flu strains, and compiling a “venom toxin peptide atlas.” Magnus warns that AI will continually discover ways around controls, so we may need AI‑driven surveillance to police AI itself. Not all experts share the alarm; researchers at Imperial College argue that current AI tools lack the sophistication to fully create bioweapons and that laboratory work remains a bottleneck. Yet infectious‑disease professor Wendy Barclay reminds us that the greatest pandemic risk still stems from naturally circulating pathogens such as H5N1, recently detected in U.S. dairy cattle and Utah mink farms. Sabra urges governments to fortify health systems, stockpile antidotes, and develop rapid‑response capabilities now, long before a crisis erupts. MIT biologist Kevin Esvelt, who pioneered gene‑drive containment methods, posted on X that a large‑language model had revealed a novel bioweapon concept he hadn’t imagined, pleading, “Please, for the love of children, the future of humanity, let’s err on the side of caution.” The debate intensifies as the world grapples with the shadow of a new pandemic scenario, and policymakers scramble to keep the double‑edged sword of AI from cutting both ways. Reuters continues to track regulatory responses worldwide.

Intel provided by Julian Reed (Consumer Electronics Expert).