The turbulent AI era is here, Bill Gates wrote on 26 August 2026, and the choices we make now are critical. That was the title he gave his essay when he reposted it on LinkedIn, a slightly sharper version of the Gates Notes headline, and it framed six thousand words as a list of decisions rather than a forecast. Four weeks later, most of those decisions have had a first answer from somebody: a president, a Congress, a minister, a foundation and a coalition of AI labs.
This article is not a second summary of the essay. We covered what Gates wrote, his five crossed thresholds and his three proposals in Bill Gates Is Deeply Worried About AI on the day it was published. The question here is what has happened since, choice by choice, in the turbulent AI era he described.
The short version is that the answers so far pull in opposite directions. Washington has rejected new guardrails while quietly proposing an AI incident channel with Beijing. Congress has a token tax bill that Gates did not write but clearly helped. Australia has rejected the idea outright. And the Gates Foundation has put a billion dollars behind the optimistic half of his argument.
Table of contents
- What Gates Meant by the Turbulent AI Era
- A Month of Answers: The Turbulent AI Era Scorecard
- The Washington Answer to the Turbulent AI Era: “Whoever Wins AI, Wins”
- The First Turbulent AI Era Institution: A US-China Incident Channel
- Taxing Tokens in the Turbulent AI Era
- The Case Against a Turbulent AI Era Token Tax
- Human Reserved in the Turbulent AI Era: Still an Idea, Not a Policy
- The Equaliser Side of the Turbulent AI Era: A $1 Billion Bet
- Where Gates Diverges From the Labs on the Turbulent AI Era
- What the Turbulent AI Era Means for UK Businesses
- Frequently Asked Questions About the Turbulent AI Era
- References
What Gates Meant by the Turbulent AI Era
Before scoring the answers, it is worth being exact about the questions, because the essay is often reduced to its most quotable line.
The essay and its two titles
The piece appeared on Gates Notes and on LinkedIn on the same morning. The LinkedIn version carried the headline most people now quote, and its subtitle states the thesis in one line: “We need a plan to ensure that the good outweighs the bad.” Gates opens with three claims: the transition to the AI era “will be one of the most turbulent times in human history”, the world is “not preparing adequately”, and the right steps would leave “everyone better off.”
The three risks he named
Gates grouped his worries into three risks. Many jobs will disappear for good, starting with entry-level and mid-level white-collar work. The technology will let people, and perhaps AI systems themselves, do far more harm, from fraud and cyberattacks to bioterrorism. And AI companions could stunt children’s development and replace human relationships.
The three proposals he made
He then offered three ideas “to start”. Build a new domestic and international system for managing the transition. Set aside some jobs for people, a domain he called Human Reserved. And rebalance how labour and capital are taxed, beginning with a tax on AI tokens and robots.
Why choices is the operative word
The essay never says the turbulent AI era can be avoided. Gates writes that he would “likely support” a credible plan to slow AI globally, then says he does not think one will happen. What he argues is that the size of the damage depends on choices made now. That makes the last four weeks a useful early test, because each choice he named now has a public response.
A Month of Answers: The Turbulent AI Era Scorecard
The table below matches each of Gates’s choices to the most significant public response since 26 August. None of these responses is final. Each one shows which way a real decision-maker is leaning in the turbulent AI era.
| Choice Gates named | What he proposed | Main response since 26 August | Direction |
|---|---|---|---|
| Institutions | National bodies plus an international AI organisation | US proposes a US-China AI incident notification mechanism (20 September) | Small step towards |
| Guardrails | Plan before disruption forces crisis mode | Trump rejects new guardrails and forms an “AI Force” (14 and 19 September) | Away |
| Taxation | Tax AI tokens and robots | House bill taxing tokens or AI revenue gains attention; Australia rejects the idea | Split |
| Human Reserved | Reserve some work for people | Wide coverage, no government adoption | Stalled |
| Benefits | Make AI an equaliser, not a divider | Gates Foundation commits $1 billion; 60-organisation language coalition | Towards |
The pace of the turbulent AI era reaction
Gates warned that “we do not have the luxury of moving slowly.” By the standard of policy, four weeks of movement is fast. By the standard of the technology, it is not: two frontier labs shipped cheaper, more capable models in the same period, as we reported in Anthropic, OpenAI Release Cheaper AI Even as Safety Fears Grow.
Who has answered, and who has not
The loudest answers have come from the United States and Australia, with smaller responses from Beijing. The European Union, the United Kingdom and most of the global south have not responded to the essay directly. That matters, because Gates’s international framework assumes that the countries that “host the leading AI developers and control critical parts of the supply chain” start meeting first.
The Washington Answer to the Turbulent AI Era: "Whoever Wins AI, Wins"
The most direct response to Gates came from the White House, and it rejected most of what he asked for.
The 14 September post
On 14 September, President Donald Trump wrote on Truth Social that there was a “SICK conspiracy going on against AI and Data Centres”. He argued that further restrictions would help China, said his administration already had substantial criminal and regulatory authority over AI companies, and wrote that the only guardrail AI needed was a “STRONG AND SMART (High IQ!) PRESIDENT”. It also included the line that has since been quoted everywhere: “WHOEVER WINS AI, WINS!”
The AI Force and its czar
Five days later, Trump announced an AI task force he called the “AI Force”, with a “czar” to lead it. “We will not in any way hinder or stifle the Growth of this incredible Industry,” he wrote, while adding that the government would be “looking for BAD” through the existing criminal and civil justice system. He also claimed AI could become “as much as 25 percent of our Country’s GDP.” Our analysis of his super intelligence rebrand at the UN covers the same week.
Gates on CBS Mornings
Gates answered on CBS Mornings on 21 September. “Most of society is still not aware of how quickly it’s moving,” he said, naming bioterrorism and financial attacks as legitimate concerns. He was careful not to attack the president: “This is not someone who has a fixed opinion. He takes in data.” He added that “if he wants to get credit for helping the entire world, the AI topic is his best bet.”
Why this matters in a turbulent AI era
Gates’s plan depends on governments acting before a crisis. The Trump position is that existing law is enough and speed is the priority. For businesses, the practical result is that federal AI rules in the US are unlikely to tighten soon, while state rules, liability claims and foreign regulation keep moving. That leaves most of the preparation Gates asked for to organisations themselves, at least in the turbulent AI era’s first phase.
The First Turbulent AI Era Institution: A US-China Incident Channel
The White House’s rejection of guardrails did not stop it proposing something that looks like the first brick of the international system Gates described.
What Bessent announced
On Sunday 20 September, after talks with Chinese Vice Premier He Lifeng in New York, Treasury Secretary Scott Bessent said the two sides had discussed a “notification mechanism” for AI incidents that could affect national security. “Moving from opaque to more transparency between the number one and the number two AI powers in the world is very important,” he told reporters. China’s state news agency Xinhua confirmed only that the talks had covered AI.
Today’s White House summit
President Xi Jinping arrived in Washington on Wednesday for a summit at the White House on Thursday 24 September, with AI, tariffs and supply chains on the agenda. The tariff truce between the two countries expires on 10 November. On Wednesday, 17 Democratic senators wrote to Trump urging him to seek an AI agreement with China, another sign that the question Gates raised is now on the table in the turbulent AI era.
How it compares with what Gates asked for
Gates said a global AI body would need elements of the nuclear inspections regime, international aviation rules and the ozone agreements. An incident hotline is much smaller than any of those. It is closest to the crisis communication lines between nuclear powers: it does not limit what either side builds, but it shortens the time before each side knows something has gone wrong.
What is still missing
No details have been published on which incidents would trigger a notification, who would send it or how fast. China has not publicly accepted the idea. Gates’s own essay said “some cooperation between the U.S. and China will be required”, so even a basic channel counts as progress on his terms. It is still a long way from the shared norms he wanted the leading AI countries to agree before competition made cooperation harder.
Taxing Tokens in the Turbulent AI Era
Of Gates’s three proposals, the token tax has moved furthest, although he did not start that process and cannot control where it goes.
The AI Tax and Work Protection Act
In early August, before the essay, Representatives Greg Casar, Valerie Foushee and Sara Jacobs introduced the AI Tax and Work Protection Act in the House. It would tax companies that develop large foundation models, sell access to them or modify open-weight models. The tax is the greater of 2% of the value of tokens processed or 3% of revenue from AI services, while unemployment is at or below 5%. The rates rise automatically as unemployment rises.
Where the money would go
The revenue would fund jobs in housing construction, infrastructure, child and elder care, education, healthcare, scientific research and local journalism. Casar has compared it to the New Deal’s Works Progress Administration. “If Congress does nothing, the rise of AI could create the biggest wealth transfer in history from the bottom to the top,” Jacobs said. Gates’s essay gave the bill a much larger audience than it had in August.
The wider Congressional menu
The House bill is one of several proposals. They differ mainly in what they tax and whether the tax is ongoing, one-off or a credit.
| Proposal | Sponsor | What it taxes or credits |
|---|---|---|
| AI Tax and Work Protection Act | Reps. Casar, Foushee, Jacobs | 2% of token value or 3% of AI revenue, rising with unemployment |
| Data centre excise tax | Sen. Ron Wyden | Changes tax treatment of AI data centres, adds an excise tax |
| Energy-based AI tax | Sen. Elizabeth Warren | Taxes AI firms partly on data centre energy use |
| American AI Sovereign Wealth Fund Act | Sen. Bernie Sanders | One-time 50% tax on OpenAI, Anthropic and xAI, with Americans given a stake through shares |
| AI retraining credit | Reps. Gottheimer and Lawler | 30% credit on AI-training costs, up to $2,500 per employee a year |
Industry voices on the rate
Two industry figures have put numbers on the idea. Anthropic chief executive Dario Amodei has suggested a 3% tax on revenue from model usage, calling it a “reasonable solution” even though it was not in his company’s economic interest. DuckDuckGo founder Gabriel Weinberg said his company would be willing to pay a 10% tax on AI token usage. Amodei’s wider call to slow down is in our piece on pacing the frontier.
The proposed rates cluster low, with one outlier. The chart compares the headline rates on the table, at unemployment of 5% or below.
The Case Against a Turbulent AI Era Token Tax
The token tax has also drawn the sharpest criticism of anything in the essay, from a government and from economists.
Australia’s “loser narrative”
Speaking at the Australian Strategic Policy Institute’s Sydney Dialogue, Australian minister Tim Ayres rejected Gates’s proposal. He said big tech firms often have an interest in arguing that their products will inevitably cause mass unemployment. “I do worry that there’s a kind of loser narrative in that proposition that suits some interests and not others,” he said. His government’s answer, he added, was “more jobs, more growth, more Australian capability.”
Token volume is not job loss
A 17 September analysis from Weiss Ratings made the technical case against. Token volume, it argued, “measures computational activity, not job destruction.” A lawyer might use a few thousand tokens to save several hours, while a scientist could use millions screening research without removing a single job. The tax would fall hardest on small firms renting AI, while the largest platforms could absorb it.
The language penalty
The same analysis cited a 2026 study of 10 models and 25 European languages that found a roughly 2.5-fold difference in tokens needed per word between English and some other languages. A flat per-token tax would therefore charge a Finnish or Hungarian user more than an English one for the same work, which cuts against the equity goal at the centre of the turbulent AI era essay.
What the IMF prefers
The International Monetary Fund has warned that a special AI tax could hold back productivity, including uses that support workers. Its alternatives are to review incentives that favour labour-displacing investment and to tax capital income and excess profits more heavily. Gates’s own underlying point about the turbulent AI era, that payroll taxes and instant write-offs “nudge” employers towards machines, is closer to the IMF view than the token mechanism suggests.
Human Reserved in the Turbulent AI Era: Still an Idea, Not a Policy
Gates’s most original idea in the turbulent AI era essay has had the most coverage and the least action.
Where the idea landed
Human Reserved takes its name from nature reserves, “places where we could put buildings and roads, but we choose not to because the loss would be too great.” Gates’s examples were care work, like the caregivers who looked after his father through Alzheimer’s, and delivering a terminal diagnosis. Moneywise, syndicated by Yahoo Finance on 23 September, framed it as a list of jobs that “should be off limits to AI”.
The turbulent AI era questions Gates left open
Gates admitted the idea “raises a host of questions I don’t have answers to.” Who decides what is reserved? What criteria apply? How do you stop companies cheating? What happens to trade when one country allows robots and another does not? No government has tried to answer any of them in the turbulent AI era’s first month. Australia’s rejection of the tax suggests reserved work will be at least as hard a sell.
The practical test for employers
The idea is still useful inside a single organisation. Every business deploying AI is already deciding which tasks stay with people, whether it writes that down or not. Our guide to human-in-the-loop AI design sets out how to make that decision explicit, which is the company-sized version of the Human Reserved question.
The Equaliser Side of the Turbulent AI Era: A $1 Billion Bet
The largest concrete response to the essay came from Gates’s own foundation. It backs the optimistic half of his argument that AI “will either be the greatest equalizer ever invented, or the worst source of injustice.”
The Goalkeepers report and the turbulent AI era
In its 2026 Goalkeepers report, published in mid-September, the Gates Foundation committed at least $1 billion over two years to wider access to AI in health, education and agriculture. “This is not a long-range prediction. It’s a present-tense choice,” Gates wrote. He added that “the decisions made in the next 12 to 18 months” would decide whether AI mainly helps people who already have the most. That is the same timescale as the turbulent AI era essay.
Education and health each take 40% of the pledge. The chart applies those published shares to the $1 billion headline.
The language gap
The report puts language at the centre of the inequality argument. More than 90% of the data used in early large language model training came from English sources, which is a turbulent AI era problem in its own right. The gap shows up in speech recognition: leading systems have error rates below 6% in English but fail more than 60% of the time in Yoruba. The report gives the example of a pregnant woman in Malawi saying her “water has broken”, which a poor model could translate as “thrown away water.”
The evidence the foundation cites
The report points to projects already running as proof that the equaliser side of the turbulent AI era is possible, not just hoped for.
| Project | Country | Reported result |
|---|---|---|
| Penda Health clinical assistant | Kenya | Diagnostic accuracy up 16 percentage points |
| Gemini Guided Learning pilot | Sierra Leone | Learning gains of up to 1.7 years in eight weeks |
| MahaVISTAAR farm advisory | India | More than 740,000 farmers, under 18 cents per person |
| Project Vaani speech data (Google) | India | Collecting 150,000+ hours of audio across every district |
The 60-organisation language coalition
At the foundation’s Goalkeepers event in New York on Monday 21 September, 60 organisations, including Anthropic, Google and the OpenAI Foundation, agreed a shared five-year goal to improve AI in under-served languages. The coalition aims to reach more than three billion people. Its governance is still being worked out, and a secretariat will track each signatory’s commitments.
The speech-recognition gap is the clearest single number in the report. The chart shows the two published error rates.
Where Gates Diverges From the Labs on the Turbulent AI Era
The Goalkeepers week also showed some tension within the turbulent AI era argument itself, and between Gates and the labs whose products define the turbulent AI era.
“We failed to build in the right human values”
On stage with commentator Van Jones at Jazz at Lincoln Center, Gates said: “We failed to build in the right human values and ways of keeping our moral code controlling what these AIs do.” The event opened with a reminder that 26 of the world’s 34 wealthiest nations cut development aid last year, which makes the foundation’s AI spending more important and harder to scale.
Full speed ahead on languages
Foundation chief executive Mark Suzman said the language work must continue “full speed ahead”, even as some of the largest AI companies call on the industry to slow frontier development. “Even if AI was frozen right now, which I don’t expect and I am not calling for, we would want to be building these language sets,” he told the Associated Press.
The labs in the coalition
The labs in the coalition admit the gap. Anthropic’s head of beneficial deployments, Elizabeth Kelly, said its products “lag in many African languages in particular.” Mozilla Data Collective chief executive E.M. Lewis-Jong put the cause more bluntly: “The internet is not a representative space.” For a business, that means an AI system bought today for a multilingual workforce may perform very differently from one language to the next.
What the Turbulent AI Era Means for UK Businesses
None of these debates is happening in Westminster yet, but each has a practical consequence for a UK organisation buying or building AI in the turbulent AI era.
Watch the tax debate, not just the technology
A token tax, a revenue tax or an energy-based levy would each change what AI services cost, and each would probably be passed on to customers. If you are signing multi-year AI contracts, check how price changes caused by new taxes or levies are handled. The US debate is early, but it is now serious enough to plan for.
Decide your own Human Reserved list
You do not need a government definition to decide which decisions in your organisation stay with a person. Write down the tasks where a human must decide, review or deliver the message, and the reason for each. Our AI strategy service builds this into deployment plans from the start.
Language coverage is a procurement question
If your staff or customers work in more than one language, test the AI system in every one of them before rollout. The Goalkeepers numbers show how large the gap can be, and vendors rarely publish per-language accuracy unless asked.
Incident reporting is coming
A US-China AI incident channel, if agreed, will make national incident reporting more likely, not less. The EU AI Act already includes serious-incident reporting duties for providers of high-risk systems. Keeping a simple log now of AI failures, near misses and customer complaints is cheap, and it will be easier to adapt than to build under a deadline.
Plan for the turbulent AI era’s 12-to-18-month window
Gates and his foundation both named the next 12 to 18 months as decisive. For most businesses that is one budget cycle, and it is the stretch of the turbulent AI era in which choices are still cheap. Use it to set governance, test vendors and train staff before the rules settle, rather than after.
Frequently Asked Questions About the Turbulent AI Era
What did Bill Gates mean by the turbulent AI era?
He meant the period of transition to widespread AI, which he expects to be “one of the most turbulent times in human history” because job losses, new harms and social disruption will arrive at the same time as the benefits.
Has anything changed since the essay?
Yes. The US has proposed an AI incident channel with China, a House token tax bill has gained attention, Australia has rejected a token tax, and the Gates Foundation has committed $1 billion to wider AI access.
Did the US government accept Gates’s proposals?
No. President Trump rejected new guardrails on 14 September and formed an “AI Force” on 19 September, although his Treasury Secretary proposed an AI incident notification mechanism with China.
What is the AI Tax and Work Protection Act?
A House bill from Representatives Casar, Foushee and Jacobs. It would tax AI developers at 2% of token value or 3% of AI revenue, whichever is greater, with rates rising as unemployment rises.
What is the Gates Foundation doing about AI?
It has committed at least $1 billion over two years, split 40% education, 40% health, 10% agriculture and 10% infrastructure and language data, and it convened a 60-organisation coalition on under-served languages.
What should a business do now?
Review AI contract terms for new taxes and levies, decide which tasks stay with people, test AI tools in every language you use, and start logging AI incidents.
References
The turbulent AI era is here. The choices we make now are critical.
A turbulent AI era and critical choices to make
After warning AI is too dangerous, Bill Gates bets a billion on its upside
Bill Gates Sounds Alarm on AI Risks After Trump Rages About Sick Conspiracy Going on Against AI
US and China discuss AI safety plan ahead of Trump-Xi summit
A new bill would tax AI tokens to fund jobs if the technology causes mass unemployment
New Democratic bill would tax AI companies to create jobs
Australian Minister Rejects Bill Gates AI Token Tax, Warns Against Loser Narrative
Bill Gates AI Tax Would Protect Big Tech, Not Workers
Bill Gates pushes for smart use of AI as his foundation builds more representative language data
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
