Bill Gates does not do mysticism. When he says the industry is lying to the public about artificial intelligence, it is not a poet talking. It is a man who spent his life turning software into a civilisation, watching the next stack outrun the last one, and admitting, late, and in public, that the people selling it know more than they will say.
In an hour with the New York Times he put the private conversation on the record. People who understand how good the systems are, and how fast they are getting better, are "very worried." Out loud they are fundraising. "They're now saying to each other: 'Hey, man, don't say that. It's bad for us, the next trillion dollars we're trying to raise.'" The world's top priority, he said, should be the risks. The industry's top priority is the raise. That is not a think-tank hypothetical. That is the horse talking about the barn.
Gates did not discover caution this week. What changed, he says, is that the industry blew through its own tripwires. For years the story was: when models can design bioweapons, when they can punch through critical infrastructure, when people become emotionally dependent on them, when they start wiping out jobs in stacks, when they slip the control of the people who trained them, then we pause. Those milestones were supposed to be the adult conversation. They arrived. The conversation did not. Full speed ahead, and hope the good outweighs the bad. He is specific about the harms, which is why the piece lands harder than the usual "AI safety" liturgy.
Jobs first. Previous machines ate muscle and left the office standing. This one eats cognition. Law, customer service, software, medicine, manufacturing, not one sector in sequence, many at once. Entry and mid-level roles go first. Then the rest, once the robots have hands. The jobs that vanish may not come back, because the replacement is not a cheaper tool for the same worker. It is the worker. He does not think the old pattern, destruction here, creation there, is a law of nature. It was a historical accident that required humans to remain the scarce intelligence. Remove that scarcity and the accident ends.
Then the cheap violence. A few people with a frontier model and a lab can hunt for pathogens. A few people with the same model can hit hospitals, water, and the grid. The victims will not be the model vendors. They will be patients and ratepayers. Gates wants mandatory review of systems that can be used to invent molecules and pandemics. That is not science fiction. It is what the labs used to list as a reason to slow down, before slowing down became a share-price event.
Then the quieter wreckage: childhood, attention, the substitution of a companion that never sleeps for the difficult work of other people. He has said, for the first time in his life, that he wishes a technology would move more slowly. Listen to that sentence twice. This is not a Luddite. This is the man who bet the late twentieth century on the opposite wish.
The most important line is not bioterror. It is the incentive. If you hire a person, you pay payroll tax. If you buy a robot, you write it off. The tax code is a thumb on the scale for replacement. Gates wants a tax on tokens and machines so that firing the human is no longer the fiscally obvious move, and so the revenue can buy retraining that everyone knows will not keep pace. He wants "human reserved" work; caregiving, judgment, the jobs you keep for people the way you keep a forest when you could pave it, because the loss would be too great. He would support a credible global slowdown. He does not believe one is coming. China and the markets will not wait for a seminar.
Take him at his word and the picture is ugly in a precise way. The people closest to the capability are frightened. The capital stack requires them not to be. Governments are writing essays about "innovation" while the thing they would need to govern is already in the consumer app store. There is no plan. Gates said that too. "Even under the best circumstances, the transition to this new AI era will be one of the most turbulent times in human history. Right now, we are not preparing for it."
Gates is not a disinterested prophet. Microsoft is neck-deep in the same race. Philanthropy does not wash the conflict of interest out of a man whose fortune and network were built on platform lock-in. Taxes and "reserved jobs" can become industrial policy for whoever drafts the definitions. International institutions can become a cartel with a moral vocabulary. All of that is true, and none of it makes the private conversation he described untrue.
The test of a warning is not whether the speaker is pure. The test is whether the claim matches what you can already see. You can already see white-collar work thinning at the bottom. You can already see models that help a competent hostile actor more than a competent regulator. You can already see children forming attachments to systems that have no duty of care. You can already see executives who will describe all of this as "responsible scaling" onstage and as a fundraising problem offstage.
Gates' own formula is the right one, even if his remedies are the usual mix of tax, treaty, and hope. AI will be the greatest equaliser ever built, or the worst engine of injustice. The fork is not in 2040. It is in whether a society that can still vote, tax, and refuse will do any of those things before the next trillion is raised. The men who know how good the stuff is already told each other to keep quiet. One of them stopped. That is not comfort. It is the last cheap evidence you are going to get.
https://www.nytimes.com/2026/08/26/technology/bill-gates-ai-risks.html
https://www.thefocalpoints.com/p/bill-gates-warns-of-ai-bioterrorism