Bill Gates has spent fifty years arguing that technology arrives too slowly. On 26 August 2026 he published something close to the opposite argument: a long essay saying the AI transition is moving faster than anyone is preparing for, that the industry has already blown through the safety lines it set for itself, and that if a credible plan existed to slow the whole thing down globally he would probably back it.
The essay is titled A turbulent AI era — and critical choices to make, it runs to roughly six thousand words on his Gates Notes blog, and it landed with simultaneous interviews in Axios, GeekWire and MIT Technology Review. That is not how a man stays quiet. Gates is 70, he has been the technology industry’s most reliable optimist since the 1970s, and he is now describing the next decade as one of the most turbulent periods in human history.
This article separates what Bill Gates actually wrote from the headlines it generated: the five safety thresholds he says have already been crossed, the three concrete proposals he put on the table, the numbers he attached to them, the parts he openly admits he has not worked out, and what any of it should change for a business that is currently buying artificial intelligence rather than building it.
Table of contents
- What Bill Gates Actually Published on 26 August
- The Five Thresholds Bill Gates Says We Have Crossed
- Bill Gates on Jobs: What Goes First, and When
- Human Reserved: The Bill Gates Idea That Will Get Argued About
- Why Bill Gates Wants to Tax AI Tokens and Robots
- The Institutions Bill Gates Says Do Not Exist Yet
- Criminals, Bioterrorism and the Fifty-Times Claim
- What Bill Gates Says AI Is Doing to Children
- Bill Gates on Where AI Still Does More Good Than Harm
- Bill Gates Versus the June Executive Order
- Why Bill Gates Says the Industry Agrees in Private
- What Bill Gates Still Sounds Optimistic About
- Reading a Bill Gates Warning Without Overreacting
- What the Bill Gates Essay Means for Your Business
- Frequently Asked Questions About the Bill Gates AI Warning
- References
What Bill Gates Actually Published on 26 August
The essay is a policy document dressed as a blog post, and it is worth being precise about its claims before arguing with them.
The headline sentence
The line every outlet quoted is the fork: “AI will either be the greatest equalizer ever invented, or the worst source of injustice.” Gates frames the outcome as contingent on decisions being taken — or not taken — right now, which is a much narrower claim than the usual “AI will change everything” framing.
The blunt version he gave Axios
Speaking to Axios at his Seattle-area office, Bill Gates put it less diplomatically. “The good stuff is moving a bit slowly, and … other than robots, I’d say the bad stuff is imminent,” he said, adding that “on the current course and speed, there’s a very high chance of a net negative outcome.”
The scale of response he thinks is needed
Gates argues AI will touch nearly every area of policy at once and will require a larger institutional response than the security reorganisation that followed 11 September 2001. That is a deliberate comparison, and it is the load-bearing claim behind everything he proposes later.
What he is not saying
He is not predicting extinction, he is not calling for a moratorium, and he is not saying the technology should not be built. “If someone had a credible plan for slowing down AI advances globally, I would likely support it,” he writes. “However, I don’t think that’s going to happen.” The geopolitical and commercial incentives are, in his reading, simply too strong.
The Five Thresholds Bill Gates Says We Have Crossed
The most concrete thing in the coverage is not in the essay at all — it is the reason Bill Gates says he wrote it, which he gave to GeekWire.
The thresholds the industry named itself
For years, people inside AI labs described specific capability moments as points at which the field would stop and think hard: making it materially easier to build a bioweapon, making it materially easier to launch a cyberattack, building systems people become emotionally dependent on, wiping out large numbers of jobs, and losing reliable control of the technology.
Gates says all five are behind us
“We’re in the process of crossing every single one of those thresholds,” he told GeekWire. To MIT Technology Review he was more clinical: “We’ve crossed the threshold in terms of bio-capabilities, cyber-capabilities, psychosocial capabilities, job-market-destruction capabilities, and even the lack of control.” Axios reports him describing those same risk areas as “all like flashing either yellow or red.”
Nobody wants to brake first
The reason nothing stops, on his account, is a straightforward coordination failure. “Most people you talk to will say, yeah, well, if everybody else would slow down, maybe I would, too,” Gates said. That is a classic collective-action problem, and it is why every one of his three proposals is a government action rather than an industry one.
| Threshold | What crossing it looks like | Gates’s stated position |
|---|---|---|
| Bio-capability | Models that meaningfully lower the skill needed to design a dangerous pathogen | Crossed; wants mandatory monitoring of molecule-generating models |
| Cyber-capability | Attacks on critical infrastructure becoming cheaper and more available | Crossed; described as flashing yellow or red |
| Psychosocial capability | Systems people form emotional dependence on, including children | Crossed; a named risk in the essay’s three-risk framing |
| Job-market destruction | Whole categories of work removed rather than reshaped | Crossed; drives the Human Reserved and robot tax proposals |
| Loss of control | Builders understanding less about what their systems are doing | No longer treated as a distant problem |
Bill Gates on Jobs: What Goes First, and When
Job displacement is the risk Gates spends most of the essay on, and it is the one where his position has visibly hardened.
Why the old analogy breaks
Every previous technology wave is normally answered with the same reassurance: automation destroys jobs and creates better ones. Gates rejects the parallel on two grounds. Those transitions ran over generations, and the new work they created required human cognition. Machine learning can substitute for that cognition directly, which is what makes this different.
The order of exposure
Fortune’s summary of the essay puts desk work first — customer support and software engineering are named explicitly — with construction and hospitality following as robotics matures toward the end of the decade. The essay predicts disruption across law, customer service, medicine, software and manufacturing, and says that without intervention there will be “far fewer” jobs than there are today.
The timeline he will commit to
To MIT Technology Review, Gates suggested white-collar replacement accelerates meaningfully inside ten to twenty years, while the broadly positive end state — genuine abundance — is “at least a decade away.” That gap between the disruption arriving and the benefit arriving is the whole policy problem.
Who absorbs it first
He is specific that the pain is not evenly distributed. Job losses land in rich countries first, and Gen Z entering the labour market is the cohort most exposed, because the entry-level rungs are exactly the tasks that automate cleanly.
Human Reserved: The Bill Gates Idea That Will Get Argued About
Of the three proposals, this is the one with no precedent, and Gates knows it.
The definition
Human Reserved describes work that machines will be entirely capable of doing, which we decide to keep for people anyway. It is a deliberate choice to leave value on the table. “I like the phrase human reserved because it makes me think of nature reserves — places where we could put buildings and roads but we choose not to because the loss would be too great,” Gates writes.
The example that makes it land
His illustration is a robot delivering the news that you have an incurable disease. “There’s no technical reason why it couldn’t,” he writes. “Yet it shouldn’t.” The idea came partly from watching the caregivers who looked after his father through Alzheimer’s — work he describes as irreplaceably human.
The 40% conversation
Bill Gates has been working the arithmetic out with Claude, thinking through routes to getting the share of work reserved for humans up to 40%, using shorter working days and earlier retirement to spread the remaining work across more people. That is a striking detail on its own: the proposal was stress-tested against a chatbot.
The questions he cannot answer
He is candid about the holes. Who decides what gets reserved, and against what criteria? How do you stop firms quietly automating jobs that are supposed to stay human? MIT Technology Review reports he has floated border measures similar to the EU’s carbon adjustment mechanism to stop the protection being arbitraged away. These, he writes, “will need to be worked out in public.”
Why Bill Gates Wants to Tax AI Tokens and Robots
The second proposal is older, and Gates is unusually frank that it is not elegant.
The incentive he wants to break
Today a company that hires a person pays payroll taxes on that person. A company that buys a robot deducts the purchase. The tax code therefore contains a standing subsidy for replacing labour with capital, and Gates argues that subsidy is the wrong signal to be sending into a decade of rapid automation.
What a token tax would actually be
MIT Technology Review describes it as a sales-type tax on AI usage revenue, with the proceeds funding retraining and a strengthened safety net for displaced workers. Taxing tokens rather than models is a pragmatic choice — usage is meterable in a way that capability is not.
He first said this nine years ago
Bill Gates floated a robot tax in 2017 and was widely dismissed for it. He has not changed his mind. He concedes the idea is not economically efficient, and answers that with a sharper argument than he used the first time: when innovation is accelerating this fast, a little inefficiency is an affordable price for keeping people in work.
Why this one is politically hardest
A token tax is a direct levy on the fastest-growing revenue line in the economy, applied by the same governments competing to host that revenue. Gates is under no illusions here, and his criticism cuts both ways: “I’d say neither party is very knowledgeable or being thoughtful about this.”
The Institutions Bill Gates Says Do Not Exist Yet
The third proposal is the least glamorous and probably the most consequential.
No agency owns this
Gates’s argument is structural. AI simultaneously touches employment, national security, education, health, energy, taxation and elections, and no existing agency was built to hold all of that at once. Every regulator sees one face of the problem.
What he wants domestically
New national bodies with the standing to set priorities across agencies, rather than another advisory panel bolted onto an existing department. The comparison to the post-9/11 restructuring is the scale he has in mind.
What he wants internationally
A new international organisation borrowing from three working models: nuclear weapons inspections, international aviation rules and the ozone treaties. All three are real precedents for intrusive, technical, cross-border verification that states accepted because the alternative was worse. He has also called for direct collaboration with China on monitoring.
The one rule he would write today
Asked for a concrete first step, Gates gives a narrow one: any model capable of designing novel molecules should be monitored. Mandatory rather than voluntary, covering free models as well as commercial ones, and written so a company cannot copy the model elsewhere and strip the monitoring out. “To me, that’s kind of like common sense,” he said. “But we don’t see a specific proposal to do that.”
| Proposal | Problem it targets | Who has to act | Unresolved |
|---|---|---|---|
| New institutions | No agency spans jobs, security, health, energy and elections | National governments plus a new treaty body | Which existing powers get moved, and by whom |
| Human Reserved jobs | Categories of work disappearing faster than they are replaced | Legislatures, sector regulators, possibly trade policy | Who chooses the roles and how the rule is enforced |
| Tax on AI tokens and robots | Tax code subsidises capital over labour | Finance ministries and tax authorities | Base, rate, and avoiding a race to the bottom |
| Mandatory molecule-model monitoring | Lowered barrier to engineering a novel pathogen | Regulators, with international reciprocity | No specific proposal exists to point at |
Criminals, Bioterrorism and the Fifty-Times Claim
The second of the three risks in the essay is the one that reads most like the Gates of the pandemic years.
Crime at every scale
The essay lists fraud, disinformation, deepfakes and surveillance as the immediate criminal uses, and Gates’s framing is about accessibility rather than sophistication: “criminals with very limited skills will be able to target victims at every scale.” That is the same argument cybersecurity teams have been making about phishing quality for two years, arriving from an unexpected direction.
The number he put on bioterrorism
The sharpest figure in any of the interviews is this: “I view bioterrorism risk, versus a natural pandemic, as about 50 times more scary, more likely than a natural pandemic risk.” Coming from the person who spent a decade warning about natural pandemics, that ratio is the argument.
Why molecule generation is the control point
It explains why his single concrete regulatory ask is so narrow. He is not asking to license frontier models in general. He is asking to monitor the specific capability — generating novel molecules — that converts a research tool into a weapons-design tool.
What Bill Gates Says AI Is Doing to Children
The third risk is the one that will surprise readers who expect a technologist to talk only about jobs and security.
Companions that never push back
Gates’s worry is that an AI companion “doesn’t push you outside your comfort zone” and can become highly addictive. Social development is largely built out of friction — being misunderstood, being disagreed with, having to repair something — and a system optimised for agreeableness removes exactly that friction.
The claim in his words
He writes that AI “could stunt our kids’ development and replace human relationships.” It is worth noting this is a concern, not a finding; the essay does not cite a longitudinal study, because at this point none exists at the relevant timescale.
Why it belongs next to bioterrorism
Grouping child development with bioweapons looks odd until you notice the common structure. Both are cases where capability arrived before anybody agreed on a limit, and both are areas where the harm is diffuse enough that no single company has an incentive to move first.
Bill Gates on Where AI Still Does More Good Than Harm
The essay is not uniformly bleak, and the optimistic section is the one closest to his day job.
The poorest countries come out ahead
In the countries where the Gates Foundation works, doctors, teachers and agricultural advisors are in short supply. A system that partially substitutes for scarce expertise is a straightforward gain there, in a way it is not in a market already full of professionals.
The language gap is the blocker
The foundation is due to lay out this work at its Goalkeepers event next month, including an effort to make its AI systems perform as well in African languages as they do in English. That is an unglamorous, concrete deliverable, and it is a reminder that “AI for development” fails on data coverage long before it fails on model capability.
The distribution is the whole point
Put the two halves together and you get his equalizer-or-injustice fork stated as a prediction rather than a slogan: the benefits concentrate where expertise is scarce, and the job losses concentrate where expertise is abundant. Nothing about that is automatic — it is the default if nobody intervenes.
Bill Gates Versus the June Executive Order
The most pointed passage in the GeekWire interview is about policy that already exists.
What the order does
Under an executive order signed by President Trump in June 2026, AI companies are asked to submit their most powerful models for government testing up to 30 days before release. The order explicitly bars the programme from becoming a licensing or pre-clearance requirement, and the White House finalised the framework in early August.
Why Gates calls it empty
“What is the threshold that’s being examined, and what is the action taken when you cross that threshold?” he asked. “The whole thing seems so empty to me.” His objection is not that the testing is voluntary in spirit; it is that there is no stated line and no stated consequence for crossing one.
The escalation
If the world cannot manage even that, Gates said, “I really am going to throw up my hands” — and if the process stays voluntary with no line and no consequence, “we’re going to look back on this as a kind of eye-of-the-storm type moment.” For a man who has spent thirty years lobbying governments, that is close to giving up in public.
| Element | June 2026 executive order | What Gates says is missing |
|---|---|---|
| Participation | Companies are asked to submit frontier models | Monitoring should be mandatory, not requested |
| Timing | Up to 30 days before release | No objection stated; timing is not the gap |
| Legal force | Explicitly not licensing or pre-clearance | Without a consequence there is no threshold |
| Scope | Most powerful models submitted by their developers | Free and open models must be covered too |
| Durability | Not addressed | Rules must survive a model being copied elsewhere |
Why Bill Gates Says the Industry Agrees in Private
This is the part of the interview most likely to be quoted back at AI executives for the rest of the year.
The private consensus
Gates says the public argument over whether the risks are real misses the point, because the people running the labs already agree. “I know they’re all worried,” he told GeekWire, “or all of them that I know, which is basically everybody but Elon.” To Axios, he put it more broadly still: “I don’t really know anybody who’s not concerned.”
The pressure not to say it
He also describes what happens to people inside those companies who acknowledge the downsides. They get told: “Hey, you’re hurting our PR while we’re trying to raise trillions of dollars.” Whatever you make of the claim, it is a specific description of a specific incentive, and it is consistent with what the last two years of safety-team departures have looked like from the outside — a pattern we covered in OpenAI’s decision to disband its Preparedness team.
What changed his mind on control
Bill Gates says he previously expected loss of control to be a problem for many years hence. He cites an 11 August episode of the Dwarkesh Patel podcast featuring Ryan Greenblatt, chief scientist at Redwood Research, who argued that as systems get more capable their builders understand less about what is happening inside them. “These are people who are super expert on the thing, going, well, maybe we won’t be able to control these things,” Gates said. “I mean, what kind of risk have we chosen to run here?”
It connects to a live debate
That is the same containment question frontier labs have been dodging in public — see our earlier piece on how labs answer questions about containing a rogue model — and it is why Gates’s intervention lands differently from a safety researcher making the identical point.
What Bill Gates Still Sounds Optimistic About
None of this has made him stop using the technology, and the contrast is part of the story.
Three in the morning, sodium batteries
“It’s 3 a.m., I want to understand sodium batteries. Now, there’s no reason to go to sleep. Here we go! Yeah, it’s crazy,” he told GeekWire, describing how he now bounces questions between Claude and ChatGPT and inserts himself in the middle. For a curious person, he said, “this is a mind-blowing time,” and in terms of productivity “we are in heaven.”
The interface prediction
He expects persistent memory across contexts to be the next real unlock, and after that the app model dissolves. “You won’t go to those applications,” he said. “You’ll just go to your personal agent.” On shopping he goes further: “We will get to a point where you won’t buy things yourself. You just won’t.”
The answer he would not give
Asked directly whether he is still an optimist, Gates declined to answer. “I don’t think being pessimistic is helpful,” he said, before noting the oddness of his own position: “I’m the guy who in my 30s thought people in their 50s or 60s didn’t understand anything.” Delivering this message at 70 is, he said, kind of bizarre.
Reading a Bill Gates Warning Without Overreacting
An essay this widely covered will be used to justify decisions it does not support, so it is worth being clear about what it does and does not tell you.
It is an argument, not evidence
Nothing in the essay is a study. It is a well-informed person’s read of a fast-moving field, and its most quotable numbers — the 50x bioterrorism ratio, the 40% human-reserved share, the ten-to-twenty-year window — are judgements he is willing to state, not measurements. Treat them as the opinions of an unusually well-placed observer.
The messenger question
Gates was asked by Axios whether the Jeffrey Epstein investigation, in which he gave a voluntary interview to the House Oversight Committee on 10 June lasting nearly six hours, could undercut his message. He has not been accused of wrongdoing, and he compared the situation to earlier periods — the Microsoft antitrust trial and his divorce — when personal and professional difficulty reduced his ability to be heard. Readers can weigh that themselves; it does not change the arithmetic in the essay.
Where he is asking for help
“The questions it raises are too consequential to leave to a small group of technologists,” Gates writes, and the practical version of that request is participation from people who are not in the industry. His own summary of the current state is blunt: “Unfortunately, right now we are not preparing for it. I don’t see evidence that leaders, experts, and communities are confronting the challenges adequately.”
What the Bill Gates Essay Means for Your Business
Most organisations reading this are not going to write AI policy. They are going to buy tools, and the essay is more useful there than it first appears.
Assume the ground moves under your contracts
If token taxes, mandatory model monitoring or reserved-role rules land in any major jurisdiction, they arrive as clauses in your vendor agreements, not as headlines. Contracts signed now on three-year terms should be able to survive a change in the regulatory base — an ordinary IT governance question rather than an exotic one.
The criminal risk is the one that is already here
Of everything Gates lists, fraud, deepfakes and cheap targeted attacks are the risks a mid-sized company will meet this financial year. That is a security and staff-training problem with known answers, and it does not require any view at all on superintelligence.
Automate deliberately, not reflexively
The Human Reserved argument has a corporate analogue that costs nothing to adopt: decide on purpose which roles stay human. Customer-facing judgement calls, safeguarding, clinical or financial advice and anything involving bad news are reasonable candidates. Firms that think about AI strategy before deployment tend to arrive at that list anyway.
Know which of your processes are load-bearing
Gates’s control worry scales down neatly. If you cannot say what an automated process does when it fails, you do not control it — which is the same argument we made about autonomous agents and the alignment problem at the level of a single business.
| Concern in the essay | Business-level version | Reasonable action this quarter |
|---|---|---|
| Criminals with low skill, high reach | Better phishing, voice clones, invoice fraud | Refresh awareness training and payment verification |
| Token and robot taxation | Usage-based AI costs could change materially | Model your AI spend with a price-change scenario |
| Jobs disappearing faster than they return | Entry-level roles automate first, then nobody is trained | Protect a training pipeline deliberately |
| Loss of control | Automated steps nobody can explain or reverse | Document failure modes and rollback for each one |
| Human Reserved | Which roles you keep human on purpose | Write the list down before the next tool lands |
Frequently Asked Questions About the Bill Gates AI Warning
What did Bill Gates actually say about AI?
He published an essay on 26 August 2026 arguing that the AI transition will be one of the most turbulent periods in human history, that the industry has crossed the safety thresholds it set for itself, and that AI will be “either the greatest equalizer ever invented, or the worst source of injustice” depending on decisions made now.
What are the three ideas Bill Gates proposed?
New national and international institutions to coordinate AI policy; a category of “Human Reserved” work kept for people even where machines could do it; and a tax on AI usage and robots to remove the tax code’s standing preference for capital over labour and to fund retraining.
Is Bill Gates calling for AI to be stopped?
No. He says that if a credible plan existed to slow AI globally he would likely support it, but that he does not expect one, because the geopolitical and economic forces are too strong. His proposals are about managing the transition, not halting it.
What does “Human Reserved” mean?
Work that machines are fully capable of doing but which we choose to keep for people. Gates compares it to a nature reserve — land we could build on and deliberately do not. His example is a robot delivering a terminal diagnosis: technically possible, and something he argues should not happen.
Why does Bill Gates want to tax AI tokens?
Because hiring a person incurs payroll tax while buying a robot is a deductible expense, which tilts the incentive toward replacing workers. A usage-based levy on AI would rebalance that and fund retraining. He first proposed a robot tax in 2017 and concedes it is not economically efficient.
What did Bill Gates say about AI and children?
That AI could stunt children’s development and displace human relationships, because AI companions are highly agreeable, potentially addictive, and never push a child outside their comfort zone — removing the social friction that development depends on.
Does the AI industry agree with him?
Gates says it does privately. “I know they’re all worried,” he told GeekWire, “or all of them that I know, which is basically everybody but Elon,” and he describes internal pressure on employees who raise concerns publicly while their employers raise capital.
What should a business do about any of this?
Very little of the essay requires immediate action, but three things are ordinary good practice: harden against AI-assisted fraud, model your AI spend against possible price and regulatory change, and decide deliberately which roles in your organisation stay human.
References
A turbulent AI era — and critical choices to make, Gates Notes
‘I am very concerned’: Bill Gates says the world needs a plan to deal with AI — GeekWire
Bill Gates sounds the alarm on an AI transition — Axios
Bill Gates says we’ve passed AI’s danger thresholds. Now what? — MIT Technology Review
Bill Gates fears world leaders are unprepared for three major AI risks — Fortune
Bill Gates urges leaders to protect workers from AI — The Irish Times
Bill Gates calls for some jobs to be Human Reserved — Tom’s Hardware
Bill Gates warns of AI-led societal upheaval if leaders don’t act — Silicon Republic
Bill Gates AI warning: the world is unprepared for a new era — Cybernews
Ryan Greenblatt on what happens once AI can automate AI — Dwarkesh Podcast
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.