Why AI Isn't Likely to Wipe Out Humanity With Bioweapons

Of all the threats posed by uncontrollable artificial intelligence, scientists say death by plague ranks low.

By Kate Knibbs | Science | September 18, 2026

Introduction

As artificial intelligence systems grow more capable, a familiar apocalyptic scenario has taken hold in public debate: an unaligned or uncontrollable AI escapes human oversight, designs a catastrophic bioweapon, and unleashes a pandemic that ends civilization.

It's a compelling narrative — and one that appears regularly in policy briefs, op-eds, and congressional testimony. But according to a growing body of expert analysis, the bioweapon scenario is one of the least plausible ways AI could threaten humanity.

What the Experts Say

Researchers across biosecurity, microbiology, and AI safety have converged on a similar conclusion: the barriers to weaponizing biology are not primarily informational. They are logistical, biological, and physical — and those barriers remain stubbornly high, even with advanced AI assistance.

The reasoning runs along several lines:

• Information is not the bottleneck. The knowledge required to construct dangerous pathogens has existed in the scientific literature for decades. AI can lower the cost of acquiring information, but that information was never truly inaccessible to determined actors.
• Biology does not cooperate easily. Designing a pathogen on a computer is far removed from successfully synthesizing, stabilizing, and dispersing it in ways that cause widespread harm.
• Wet-lab constraints are real. Synthetic biology requires specialized equipment, controlled supply chains, and skilled personnel — all of which are difficult to obtain covertly and are increasingly subject to screening and regulation.
• Failures are common. Historical attempts at bioweapons development, including state-sponsored programs, have typically produced far more failure than success.

Why the Scenario Persists

The bioweapon narrative persists for understandable reasons. It is vivid, dramatic, and maps onto real historical fears about weapons of mass destruction. It also offers a simple story about AI risk in which a single, identifiable technology — a model — is the villain.

But experts caution that overly focusing on this scenario can distort risk prioritization. If the goal is to reduce real-world harm, attention may be better spent on threats that are demonstrably more feasible: cyberattacks, disinformation, surveillance, labor disruption, and the erosion of institutional trust.

A More Grounded View of AI Risk

This doesn't mean AI poses no biological risk. Lowering barriers to entry — even modestly — could expand the pool of actors capable of causing harm, and that possibility deserves serious attention from regulators and researchers alike.

The more accurate framing is one of degrees, not extremes. AI may make certain biological tasks easier, but "easier" is a long way from "capable of ending humanity." Treating the bioweapon scenario as the primary AI threat risks preparing for a threat that may never materialize while neglecting ones that already are.

The Bottom Line

As AI capabilities advance through 2026 and beyond, the conversation around existential risk will continue. But when scientists rank the most pressing dangers associated with uncontrollable AI, death by engineered plague sits near the bottom of the list — not because the threat is impossible, but because the practical barriers to pulling it off remain formidable.

Originally published in WIRED.

via Wired AI