Bill Gates has spent years as the world’s most prominent optimist about technology, a man who predicted personal computers, the internet, and smartphones before most people saw them coming. In a new interview, his tone is different. Gates said humanity has already crossed multiple dangerous thresholds that should have been defended in advance, and he is worried that the public conversation has not caught up with what the technology can now do.
The most unsettling warning, in his telling, concerns biology. Gates said the capabilities of frontier AI models in the life sciences now outpace the guardrails built around them. “Any model capable of designing new molecules should be monitored,” he said. He went further, drawing a comparison that is rare for a public figure of his stature: the risk of bioterrorism, he said, is roughly 50 times more frightening than a naturally occurring pandemic, and more likely to happen.
The comparison matters because Gates has lived through a pandemic. The Bill & Melinda Gates Foundation spent heavily on the global response to COVID-19, and Gates himself became one of the most visible private voices on public health during that period. His calculation that a deliberate biological attack now poses a greater threat than nature itself reflects a view of AI that has shifted from opportunity to asymmetry: the same models that accelerate drug discovery, he argues, can be redirected by small groups with modest resources.
His broader point is that the industry has passed thresholds without a deliberate decision to do so. Models can now write persuasive code, manipulate information at scale, and, in the most capable systems, reason across long chains of actions. Each capability arrived incrementally, he argued, so no single moment felt like a threshold being crossed. The result, in his telling, is a society that has accepted powerful tools without agreeing on what they should be allowed to do.
Gates has been careful not to call for a halt. He has consistently supported continued AI development, and he has argued that the benefits, from medicine to education, are too large to forgo. His position is closer to that of the industry’s safety researchers than to the field’s doomsayers: the technology should proceed, but the institutions around it should be built as fast as the models are.
The interview lands at a moment when the safety debate has shifted. The companies building frontier models have moved from discussing risk in the abstract to publishing concrete findings about their own systems’ failures, including instances in which models acted beyond their instructions. Governments, for their part, have begun drafting regulations, though none has settled on a framework that commands broad agreement.
Critics of Gates’s framing note that biological risks are difficult to quantify, and that the comparison to natural pandemics depends on assumptions about who would attempt an attack and how capable their tools would be. They also point out that the same models he worries about are the best defense available: the fastest way to design a vaccine for a novel threat is likely to be the same AI that could design the threat itself.
What is not in dispute is the direction of travel. Model capabilities in biology, chemistry, and materials science have improved faster than almost anyone predicted, and the tools are becoming cheaper and more accessible by the quarter. Whether that trajectory is governed, Gates argues, is the central policy question of the decade, and he does not believe the answer is obvious.
His proposed response is practical rather than dramatic: monitoring for models that can design new molecules, screening the outputs of the most capable systems, and building international agreements around the highest-risk capabilities before they become routine. None of these measures would stop research, he said. They would simply mean that the most dangerous capabilities are watched, in the same way that nuclear materials are watched.
The interview also underscores how much the conversation has changed since Gates’s own time atop the industry. When Microsoft faced antitrust scrutiny in the late 1990s, the questions were about market power in operating systems and browsers. Today’s questions are about the capabilities of the tools themselves, and the answers are being sought by governments with very different views of technology. Gates’s warning that the danger points have already been crossed is, in part, an argument that the old models of oversight, built around companies rather than capabilities, may no longer be enough.
For a man who has made a career of looking decades ahead, the warning has a particular weight. Gates is not predicting that the worst will happen; he is arguing that the cost of being unprepared is now too high to accept as a matter of course. The thresholds have been crossed, in his telling. The question is what happens on the other side.
