Bill Gates Says AI Has Already Crossed Its Danger Thresholds

Key points

• Bill Gates published a new essay in mid-2025 warning that AI has already crossed thresholds for biological, cyber, psychological, and job-market danger.
• Gates estimates that a bioterrorism threat enabled by AI is roughly 50 times more likely to cause mass harm than a natural pandemic.
• He proposes two new policy ideas: legally reserving certain jobs for humans, and taxing the use of AI tokens and robots that replace human workers.
• Gates says current economic statistics hide how fast white-collar job displacement is already happening.
• He stresses that protesting data centres is not an effective response, because AI development will continue globally regardless.

Bill Gates does not look like a man at peace. Seated in a conference room at his Kirkland, Washington offices, the 70-year-old former Microsoft chief executive was rocking back and forth as he spoke, visibly agitated. His message, delivered to MIT Technology Review in a wide-ranging interview published alongside a new essay, was blunt: society is not worried enough, and the window to act is closing fast.

What thresholds has Gates crossed off his worry list?

Gates says AI has already passed the point where it needed to be brought under control across five areas: its ability to help design dangerous biological agents, its use in cyberattacks, its influence over human psychology, its capacity to destroy jobs, and the simple question of who is actually in charge of these systems.

"I'm in a state of shock that we've crossed these thresholds," he told MIT Technology Review. For years, he says, the assumption inside the technology industry was that guardrails, meaning safety rules and technical limits, would be firmly in place before AI reached these danger points. They were not.

His sharpest warning is about biology. Any AI model capable of designing novel molecules, meaning new chemical or biological compounds that do not yet exist in nature, should be actively monitored, he says. He puts the risk bluntly: a bioterrorism attack made possible by AI is, in his estimate, about 50 times more dangerous than the threat of a natural pandemic.

Will AI really wipe out that many jobs?

Yes, he believes so, and faster than most economists expect. Gates argues that this time, unlike every previous wave of technology, AI can replace human thinking across nearly every industry at the same time, at low cost, and with error rates that will probably beat humans.

"The past is just very misleading," he said. History shows that new technology always created new jobs to replace the old ones. Gates says that argument no longer holds when a single tool can do the cognitive work of a lawyer, an analyst, and a customer service agent for a fraction of the salary.

Some companies are already quietly hiring fewer entry-level workers. He calls that a modest early signal of something much larger coming.

What does Gates actually want to happen?

Two concrete proposals stand out. First, he suggests governments agree to reserve certain jobs for human beings, categories that society decides should stay in human hands. Second, he wants a tax on AI tokens (each token is a small unit of text or data processed by an AI system) and on robots that replace workers, with the money set aside for affected communities.

He is not without hope. The Gates Foundation is using AI to speed up vaccine and drug research. He also points to a spin-off group called NextLadder, which is building AI tools to help low-income families decode government benefits, eviction notices, and bankruptcy proceedings, the kind of complex paperwork that usually requires a lawyer most people cannot afford.

But optimism comes later. First, he says, turbulence.

One thing he does not want: data centre protests. Blocking construction in the United States would not slow AI development, because the computing facilities, the large buildings packed with specialised chips that run AI systems, will simply be built elsewhere. "Going to a data centre protest is not the most effective way to start the debate," he said. The fight, in his view, needs to happen in policy rooms, not car parks.

Common questions

Should ordinary people be worried about AI right now?

Gates says yes, specifically about job losses in white-collar work over the next few years, though he also sees real benefits in health care and cutting through government red tape. The practical step is to follow policy debates about AI regulation, not assume the industry will self-police.

What is a robot tax and would it actually work?

A robot tax is a levy on companies that replace human workers with automated systems, with the money going to retraining or social support. Gates has supported the idea for years; no major government has adopted one yet, so its real-world impact remains untested.