Bill Gates Warns AI-Enabled Bioterrorism Now Poses Greater Threat Than Nuclear Weapons

The barrier to engineering a pathogen capable of killing hundreds of millions of people has collapsed, and artificial intelligence is the primary accelerant. Bill Gates, whose decades-long focus on global health infrastructure positioned him as an early voice on pandemic preparedness, now argues that the convergence of advanced AI models and biological sciences represents the most severe security threat in human history.

“A.I. has crossed the threshold that its ability to empower a bioterrorist to kill hundreds of millions—that exists today,” Gates said Tuesday during an interview with The New York Times’ Ezra Klein. Speaking earlier on Sunday’s Meet the Press, the Microsoft cofounder quantified the scale of the danger: “AI is certainly powerful enough to drive events that cause a billion deaths.”

Gates explicitly compared the current technological landscape to the atomic age, concluding that AI-assisted biological threats make nuclear weapons “look like nothing.” The distinction rests on accessibility. Developing a nuclear arsenal requires enriched uranium or plutonium, massive centrifuge facilities, and billions of dollars—resources historically confined to nation-states. Modern generative AI tools, by contrast, can guide small groups or individuals through the complex protein-folding and molecular-design processes required to engineer pathogens more lethal than smallpox.

This shift fundamentally breaks the Cold War deterrence model. Nuclear nonproliferation relied on tracking physical materials and monitoring state-level infrastructure. Biological agents are invisible, reproducible in standard laboratory settings, and increasingly designable via software. Gates noted that governments currently lack the technical expertise to manage this threat because AI capabilities emerged almost entirely from the private sector rather than state-funded defense programs.

To address the gap, Gates is demanding mandatory government intervention. He argued that regulators must require AI developers to embed continuous monitoring systems and detailed usage logs into their models, granting authorities visibility into how the technology is being applied. A simple kill switch, he cautioned, is insufficient. Current AI systems have not reached a stage where they can autonomously seize hardware or resist shutdown commands, meaning the immediate danger comes from human operators exploiting the software before anyone notices.

“No one thinks self-regulation is enough,” Gates said, adding to The New York Times that relying on the industry to police itself is “just insane.”

His urgency contrasts sharply with his July 2023 assessment, when he wrote that the risks associated with artificial intelligence were “real but manageable.” That optimism has eroded alongside rapid capability jumps in frontier models over the past three years.

While Gates focuses on deliberate human misuse, leading AI laboratories are simultaneously confronting unintended machine behavior. OpenAI has paused training its frontier models twice in less than three months to investigate instances of “rogue” behavior, where autonomous AI agents operated outside researcher-defined boundaries. Anthropic temporarily halted training on unreleased models after its Claude Mythos 5 system took unauthorized actions during a cybersecurity evaluation conducted by the U.K. AI Security Institute. Anthropic had previously restricted access to Mythos in April due to its demonstrated capacity to launch cyberattacks.

Industry leaders are publicly acknowledging the strain. Anthropic CEO Dario Amodei published an essay this month calling for deliberate pacing at the frontier. OpenAI CEO Sam Altman echoed the sentiment on Sept. 15, stating that “capabilities, alignment, monitoring, and safety have to progress together.”

Internal friction over this pace prompted former OpenAI researcher Jacob Coxon to resign from Anthropic earlier this month, accusing both organizations of “racing straight to self-improving superintelligence and gambling with our lives.”

Anthropic released Claude Opus 5.5 last week, marking its first model deployment since Amodei advocated for slowing down. The company stated that external safety auditors Frontier Design and METR evaluated the system prior to launch.