AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Bill Gates says to prepare for 20 years of ‘tumultuous’ AI disruption

Collected Oct 22, 2026

Bill Gates used a Decoder interview with The Verge to argue that society should prepare for roughly two decades of disruption from AI, and that the technology has crossed a capability threshold he now describes as the most dangerous ever invented. The conversation follows a late-August essay of his titled "The turbulent AI era is here. The choices we make now are critical," in which he wrote that even under the best circumstances the transition will be one of the most turbulent times in human history. Since then he has called AI "the most dangerous thing that humans have ever gone near" and said it "makes nuclear weapons look like nothing."

Gates framed his position as worried rather than fatalistic. He does not accept the doomer label; he calls himself a strong voice that this is the most profound technology ever, both in its capacity for good and in the pace of change it will force. He also repeated an earlier position that AI is beneficial if shaped correctly, adding that the response so far has not matched the speed of the challenge. In his account, the risk is two-sided: harms may not be minimized, and the benefits may arrive slowly, if the transition is badly managed.

What justifies the urgency, in his telling, is progress over the past year across several fronts at once: writing code and finding cyber flaws, reasoning in biology, forming relationships with children and other people, and disrupting the job market. Because those capabilities already exist, he said every one of those effects will happen. He expects the disruption to start in white collar and intellectual work and to reach physical and blue collar work fairly soon through humanoid robots. A machine confined to medicine, law or education alone would already be hugely disruptive; this one touches every area of human activity.

On near-term risk, Gates drew a line between models that keep safeguards and those where protections can be stripped away. The big near-term danger, he said, comes from models without safeguards, because that is where cyberattacks are available today. Models whose operators monitor how they are used and guard against removal of restrictions carry a much smaller near-term risk, though on a longer horizon the control problem applies to any state-of-the-art model. For a cyberattack, he noted, the choice of model is irrelevant, which is why he does not think safety can be reduced to picking a trustworthy vendor.

He was skeptical of the idea that frontier labs drifted from their founding ideals simply because of market pressure. The governance constructs at OpenAI, Anthropic and DeepMind were designed to restrain commercialization, but in his view they only work when a single company sits at the frontier. With many firms capable of building state-of-the-art systems, that arrangement breaks down regardless of whether the people involved still understand the risks. Once competition is plural, responsibility cannot rest with companies, he argued; it has to move to government, as it does for cars and planes, though he conceded that the scale of change here dwarfs those domains. He voiced concern about whether governments can understand what they would regulate, and said he has not heard anyone in the industry say they want to stop.

Gates repeatedly declined to argue about what counts as AGI or whether models are conscious. He said he does not believe current systems are conscious, but that none of the safety issues turn on that word. The question that matters is whether a tool exists that could shut down banks or electricity grids, and there is a business model behind cyberattacks that does not care about definitions. A larger concern in his framing is that these systems take dramatic harm out of the hands of nation states, and the larger ones at that, and hand it to independent groups, whether profit-seeking in cyber or destructive in bio. He called it stunning that there is still no mandatory review of models.

He sketched three approaches to the longer-term control problem. One is to surround a model with an external layer that inspects its actions against explicit symbolic constraints, such as whether sending mail to strangers or taking over another cluster was expected. Another is to change how the models are written, because anything left out of the reinforcement learning objective can be abused. A third is white-box understanding of a model's internals, so mistakes are visible rather than discovered only by watching behavior. He described today's practice as growing models and observing how they behave, and said control work is a huge priority but not the imminent threat compared with children's use of AI, the job market, bioattacks and cyberattacks.

Gates also pushed back on the idea that industry warnings are marketing or hype to support an investment bubble, saying the concerns voiced by people close to the technology are sincere and close to universal. And he observed that outside the industry, willingness to believe this time is different is very low, because past warnings about technology and jobs were not borne out.

Why it matters: Teams building on frontier models should expect safety expectations to become a regulatory matter rather than a vendor promise, with mandatory review a plausible direction if Gates's argument gains traction — though that is inference, not a stated policy. For product teams, the practical read is that safeguard design, logging of agent actions and external constraint layers may move from nice-to-have to baseline. The split-society scenario he describes also suggests product trust and adoption could fragment along lines of comfort with the technology frontier.

Read at The Verge · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Today, I’m talking to Bill Gates, cofounder and former CEO of Microsoft and chairman of the Gates Foundation. You may have noticed Bill in the news more lately — he’s been on something of a tour for the past few months, ever since publishing an essay in late August titled “The turbulent AI era is […]