Microsoft co-founder Bill Gates published a nearly 6,000-word essay on his personal website Gates Notes on Wednesday, arguing that artificial intelligence will bring about "one of the most turbulent periods in human history," and admitting that "there is currently no plan" to cushion the transition. In a separate interview from his Seattle office that same day, he stated that at the current pace, "the probability of a net negative outcome is very high."

The essay, titled "The Choices We Make About AI Right Now Matter," declares that AI will either become "the greatest equalizer ever" or "the worst source of injustice." Gates outlines three risks that are already emerging simultaneously: the permanent disappearance of a vast number of jobs, a lowering of the barrier to committing harmful acts, and the replacement of children's growth and human relationships. Decisions on government legislation, tax reform, and which jobs remain reserved for humans have yet to be finalized.

Gates has revised his earlier stance: three years ago he called the impact manageable, but now he says a significant chunk of jobs will vanish. Back then, he wrote that AI-driven job disruption would be "bumpy" but still "manageable," with technology helping people work more efficiently. Speaking to Semafor on Wednesday, he described feeling a sense of shock: "I was almost the first person to say, 'This is crazy, this is absurd.'" He told Axios, "I'm deafened by the silence." Good things are moving slowly, he said, but aside from robots, "the bad things are already imminent." Under the current trajectory, the probability of net negative outcomes is very high. He also noted that nearly every industry insider he knows is worried; adding to GeekWire, "Basically, everyone except Elon is worried."

The essay frames the job impact as a structural problem, not a cyclical unemployment wave. Gates cites the roughly 25% unemployment rate during the 1933 Great Depression as a comparison, noting that AI "may not reach that level," but the disruption "will not disappear with the economic cycle." In white-collar sectors, customer service, software engineering, and paralegal roles will be replaced first; on the blue-collar side, he writes that "many Americans don't realize how fast dexterous robots are improving," and some physical labor in construction and hospitality could begin competing with humans by the end of this decade. Without intervention, newly created jobs would be "far fewer" than today's total employment. The transition would unfold over about a decade, not generations. He told Semafor that in most professions, humans will ultimately be "far worse than AI doing the job." He questioned how large the rebound after job losses might be: "It would make the data center thing look like small potatoes."

Foundations cannot close the social security gap. He points out that government spending is roughly 20 times the size of all charitable giving; the Gates Foundation's annual budget of about $10 billion is "far from sufficient." The response cannot be "industry self-regulation."

The three risks are laid out side by side in the same essay. The first is jobs. Gates writes that entry-level and mid-tier positions are most dangerous, as new roles often require skills that take years to master. He specifically highlights young people: entry-level jobs disappear first, and the ladder for re-employment narrows. Waiting until "people have already been displaced or underemployed" would be too late.

The second risk is the lowered barrier to wrongdoing. Scams, deepfakes, disinformation, and surveillance will enter daily life first. The resource threshold for cyberattacks is dropping, with infrastructure like hospitals and power grids named as weak points. On the biological side, he writes that AI will accelerate drug and vaccine development but also make designing deadly new diseases easier—"the positive capabilities and dangerous capabilities are hard to separate." Speaking to GeekWire, he recounted the warning lines the industry has verbally crossed: easier bioweapons, easier cyberattacks, machines that create emotional dependence, the elimination of vast numbers of jobs, and technology spiraling out of control. "We are crossing every single line."

The third risk concerns children and relationships. The essay cites a study of over 1,100 people from Stanford and Carnegie Mellon: the heavier the use of AI companions, the worse the self-reported well-being. Gates writes that as a child, he put effort into learning social skills and making friends, "if I had an AI companion back then, I doubt I would have put in the same effort." He also mentions preliminary surveys showing that heavier use correlates with weaker critical thinking, especially among younger groups. He describes these risks as "yellow or red lights already flashing."

He criticizes current voluntary safety arrangements: the thresholds, the consequences for crossing them—"the whole thing looks empty to me." Without clear red lines and consequences, "looking back, we will see this moment as the eye of the storm." If someone presented a credible global slowdown plan, he writes he would "likely support it," but "I don't think it will happen." Geopolitical and political forces are too strong.

He proposes three things: new institutions, reserved human jobs, and taxing tokens and robots. First, building new institutions. Gates writes that existing bodies were not designed for a technology that simultaneously touches employment, security, health, energy, and elections. He wants a domestic agency that can set cross-departmental priorities, and an international organization drawing parts from nuclear inspection, civil aviation rules, and ozone treaties. This is a proposal, not an established body. On Axios, he assessed the two U.S. parties: "Both sides don't really understand this, and haven't seriously thought about it." To GeekWire, he asked, "Who should I talk to on the White House side?" Washington's legislative agenda remains unchanged.

Second, "Human Reserved" jobs: roles that machines can already perform, but society decides to keep for humans. The analogy is a nature reserve—land that could be built on or used for roads, but is deliberately left untouched because the loss would be too great. Examples include childcare, jury duty, and telling someone "you have an incurable disease." He writes: "There's no technical reason it can't be done. But it shouldn't be." A caregiver tending to a demented father is described as "work that cannot be replaced by machines." In education and healthcare, AI should not be banned outright, but jobs should be protected, humans remain in charge, and tools are given to people. Reservations can be two types: permanent human roles, and temporary protection for those in career transitions, such as manufacturing workers in the late stage of their careers. In extreme cases, he told Axios, one could initially imagine reserving about 40% of jobs for humans—"that's the upper limit I can think of." Which specific roles and how to implement them, the essay states, "must be publicly discussed." Progress can vary by country, phased in over years or decades, rather than replacing everything overnight. None of this is law yet.

Third, taxes. Gates advocates taxing AI tokens—the billing units models use to break down text and other data—and robots, to reduce the financial incentive to replace humans with machines and to fund retraining and safety nets. The current tax system taxes employers for hiring people, but buying robots is usually deductible as a business expense. "The tax code is pushing you to replace people with machines." He told Axios this would be "the biggest tax change I've seen in my lifetime," colliding with "the most divided politics I've seen in my lifetime." He proposed a robot tax nine years ago, which was widely shelved, and he still stands by it. He admits it's not "economically optimal," but as innovation accelerates, society can afford a small efficiency loss to keep people employed. The essay does not provide a legislative text on rates, tax base, who levies it, or how the money is distributed.

The Gates Foundation continues to spend its remaining roughly $200 billion over a 20-year plan, with 19 years left, applying AI to health, agriculture, and education programs. He also says philanthropy cannot fill the national safety net. Model companies "are not policy people" and are "a bit evasive" when discussing the negative aspects of their own products. The industry cannot solve this alone. What has been issued is a warning; what remains unimplemented is regulation.

Gates frames this essay as the first in a series, with future pieces dedicated to biological risks. The 70-year-old says that in his thirties, he thought people in their fifties and sixties didn't understand anything, and it's "a bit odd" for him to be issuing such warnings now, but "I'm very worried. One-on-one, they are too." He is willing to stake his reputation on his job market prediction: "I hope someone convinces me I'm wrong about the job market, but no one has. This will be very bad for jobs." What was issued on Wednesday is a personal initiative and a set of interview talking points: the three risks are fixed, he does not expect a slowdown plan to materialize, and the domestic agency, international body, reserved human jobs, token tax, and robot tax all remain on paper. Which jobs are reserved for humans, what the tax rate is, and who the White House liaison is—none of these have been turned into enforceable rules.