NextFin News - Bill Gates, the Microsoft co-founder who spent the past decade as one of technology's most prominent optimists, has issued his starkest warning yet: artificial intelligence is now powerful enough to "cause a billion deaths" if it falls into the hands of people with ill intent. The warning, delivered in a new interview and a sweeping 5,784-word essay, marks a decisive break from Gates' earlier enthusiasm for AI — and it comes with a concrete, if politically difficult, prescription: governments must tax AI usage and robots, reserve whole categories of jobs for humans, and build an international watchdog modeled on nuclear-weapons inspectors.
The shift matters because of who is speaking. Gates is not an outsider sounding an alarm at an industry he does not know. He co-founded the company that has invested billions of dollars in OpenAI and embedded AI across Windows, Office, and Azure, and he acknowledged in the essay that he still holds financial ties to the tech industry. When the architect of part of the AI buildout says the builders are moving too fast, the market cannot dismiss it as Luddism.
The Warning, in His Own Words
In the new interview, Gates said: "AI is certainly powerful enough to drive events that, you know, cause a billion deaths." The mechanism he describes is not a runaway superintelligence from science fiction but something more immediate: malicious actors wielding AI tools that are already available. He framed the catastrophic risk as coming from "people with ill intent using the latest AI tools," with bioweapons and cyberattacks as the most plausible vectors.
"The good stuff is moving a bit slowly, and ... other than robots, I'd say the bad stuff is imminent."
That line, from the same interview, captures the asymmetry at the heart of his argument. The benefits — drug discovery, personalized education, productivity gains — are arriving slowly and unevenly. The harms, in his view, are close.
"On the current course and speed, there's a very high chance of a net negative outcome," Gates said. The essay, published in late August under the title "The turbulent AI era is here. The choices we make now are critical," opens with a binary: "AI will either be the greatest equalizer ever invented, or the worst source of injustice." And even on the most favorable assumptions, he wrote, "the transition to this new AI era will be one of the most turbulent times in human history."
From Pandemic Prophet to AI Cassandra
Gates, 70, is not a newcomer to catastrophic-risk warnings. In a March 2015 TED talk, he said, "If anything kills over 10 million people in the next few decades, it's most likely to be a highly infectious virus rather than a war. Not missiles, but microbes." That talk became a touchstone during the COVID-19 pandemic — a case where he was early and, by most measures, right.
Now he has escalated the frame. In the new essay he writes that "an even greater risk than a naturally caused pandemic is that a non-government group will use open source AI tools to design a bioterrorism weapon." The logic is a direct extension of the 2015 argument: the world underinvested in pandemic preparedness because the threat was probabilistic and distant; the same myopia is now repeating with AI.
The reversal in tone is measurable in his own words. In an interview last October he called AI "the biggest technical thing ever in my lifetime" and said its influence was hard to overstate. By late August he was arguing that the traditional analogy to earlier technological revolutions no longer holds. Past transitions unfolded over generations and ultimately created new work requiring human cognition. AI, he argues, can substitute for that cognition itself — which is why his essay predicts "far fewer" jobs than exist today absent intervention.
"They're just full speed ahead and hoping that the good outweighs the bad," he said of the industry's posture.
The Mechanism: Bad Actors, Not Rogue Machines
The core of the "billion deaths" claim rests on a specific transmission channel: AI lowers the skill floor for causing mass harm. Designing a pathogen, engineering a toxin, or mounting a large-scale cyberattack once required state-level resources or rare expertise. Widely available models compress that gap. The risk is not that the AI decides to kill; it is that a small group, or even an individual, gains capabilities that previously belonged to governments.
The economics reinforce the danger. Training a frontier model once cost hundreds of millions of dollars and required specialized hardware — a barrier that functioned, however imperfectly, as a form of access control. That barrier is falling. Open-weight models that rival the capabilities of proprietary systems now run on consumer-grade hardware, and the performance gap between the best closed models and the best open ones has narrowed to a matter of months rather than years. Capability is diffusing faster than governance can be built.
Consider the sequence of milestones the industry itself once treated as red lines: systems that could pass professional exams, write working code, or explain how to handle dangerous pathogens. Each was described, when breached, as a reason to pause and reassess safeguards. All are now routine features of widely available tools. Gates' point is that the warning signs were not hypothetical; they were observed, acknowledged, and then crossed without a coordinated response.
This is a structural claim, not a cyclical one. A cyclical risk mean-reverts — a recession ends, a virus burns out, a stock corrects and recovers. A structural risk changes the underlying parameters of the system. Once a capability exists and its marginal cost falls toward zero, it does not un-exist. That asymmetry is what makes the framing so severe: the downside is essentially unbounded, while the upside accrues gradually and unevenly.
"The people who need the most time are the ones who have the least — the accounting worker who's replaced by a bot or the $20-an-hour worker who loses their job to a $10-an-hour robot," Gates said. "Robots and AI combined can create a vicious cycle."
The Policy Answer: Taxes, Job Reservations, and a New Global Body
Gates' prescriptions are as unconventional as his warning. He wants governments to consider:
- A tax on AI usage — what has been described as a "token tax" — and a tax on robots, to counteract a tax system that, in his words, "nudges you toward replacing people with machines."
- "Human Reserved" jobs — categories of work set aside for people. In an extreme form, he said he could imagine 40% of jobs initially reserved for humans. "But that's as high as I can get," he added.
- New national institutions to coordinate AI policy across employment, taxation, energy, elections, public health, the financial system, and national security.
- An international organization with elements of the nuclear-inspections regime, international aviation regulation, and ozone-layer agreements.
He is under no illusion about the odds. "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." On domestic politics, he was equally blunt: "I'd say neither party is very knowledgeable or being thoughtful about this."
His foundation, meanwhile, plans to spend more than $1 billion to expand equitable access to AI in education, healthcare, and agriculture — an attempt to tilt the distribution of the upside toward the poorest, even as he warns about the downside.
The Market Is Not Listening — Yet
The market reaction to Gates' warnings has been muted. Microsoft shares closed at $514.30 on Sept. 25, up 3.0%, essentially ignoring the founder's own catastrophe framing. Nvidia, the company whose chips power the AI buildout, closed at $224.07, down 0.5%. The disconnect is the story: Gates is describing a tail risk that markets, by design, do not price until it stops being a tail.
The valuation math explains the complacency. The AI investment thesis rests on a simple premise: capital expenditure on data centers and models will convert into durable earnings growth as enterprises adopt AI across workflows. That premise assumes a stable regulatory environment in which automation is taxed, if at all, no differently than any other capital investment. That is the second-order implication investors are missing: Gates' policy agenda attacks exactly that assumption. A token tax raises the marginal cost of inference; a robot tax raises the return threshold for replacing workers; "human reserved" categories cap the addressable market for automation in protected sectors. None of these would eliminate the opportunity, but each would compress the terminal margin that today's prices imply.
There is also a positioning asymmetry. The AI complex — chipmakers, cloud providers, model labs, and the utilities building their power supply — has become a dominant share of major equity indices. A coordinated regulatory shock would not be idiosyncratic; it would be systemic. That is why the market's silence is rational in the short run and potentially costly in the long run: pricing in a low-probability, high-severity regulatory event today would mean underperforming while the boom continues, but failing to price it at all means holding assets whose cash flows depend on a policy path that a single high-profile incident could overturn.
The industry is pushing back by name. Jensen Huang, the chief executive of Nvidia, responded to Gates' criticism in late August. "I see things very differently than he does," Huang said. "I love the heck out of Bill, but I don't see what he sees." On the employment question he held firm: "Overall, this is going to be a net job creator at a scale that we have never seen." That exchange — philanthropist warning versus builder accelerating — is the central tension of the AI era, and it is unlikely to be resolved by persuasion.
The Strongest Counter-Thesis
The most credible objection to Gates' position does not come from AI boosters. It comes from the historical record of technological pessimism itself. Every transformative technology — the printing press, the steam engine, electricity, the internet — produced moral panics about unemployment, social decay, and weaponization. Most of those panics proved wrong on the employment question: technology destroyed specific jobs but raised living standards and created new categories of work. The counter-thesis is that Gates is repeating the Luddite error at a grander scale, and that his "Human Reserved" proposals would freeze productivity gains that could lift global welfare.
The historical record also cuts the other way on safety. Nuclear technology, the closest analog to a dual-use capability that can end civilizations, was brought under a nonproliferation regime precisely because the downside was recognized as existential. The counter-thesis concedes biosecurity but argues the remedy is narrow — export controls on biological design tools, screening of model outputs, international pathogen surveillance — not the economy-wide intervention Gates proposes. On that view, Gates is right about the disease and wrong about the treatment: the appropriate response to a bio-risk is a biosafety regime, not a tax on every AI transaction or a mandate reserving 40% of jobs for humans.
That narrower critique is harder to refute than blanket optimism, and it deserves weight. But it still requires governance to work — and Gates' central empirical claim is that governance has consistently arrived after the capability, not before. The burden of proof, on his framing, lies with anyone who believes this time the sequence will reverse.
Gates anticipates part of the economic critique. He has said he would love to be convinced he is wrong about the job market but is not. And his bioweapon argument is not a claim about net job loss — it is a claim about a qualitatively new class of risk. The printing press did not give peasants a recipe for a pandemic. The counter-thesis is strongest on economics and weakest on biosecurity.
The falsifying signal is concrete: if, over the next three to five years, AI deployment proceeds without a single serious bioweapon or catastrophic cyber incident attributable to accessible models — and if employment-to-population ratios in advanced economies hold steady while real wages rise — then the structural-risk framing is overstated and the policy agenda should be scaled back to targeted biosecurity and cyber controls rather than economy-wide taxes and job reservations.
Who Benefits, Who Is Exposed
If Gates' warnings translate into policy, the impact is not symmetric across the AI complex. The most exposed are the companies whose valuations depend on the fastest possible automation of labor: enterprise-software vendors selling headcount-replacement workflows, robotics integrators, and any business model premised on inference volumes growing without a tax wedge. The least exposed are the picks-and-shovels suppliers of compute and power, whose demand is driven by training and inference regardless of who pays the tax — though even they would feel a slowdown if regulation throttles deployment.
The potential beneficiaries are the firms selling compliance, safety tooling, and audit infrastructure — the corporate analog of the nuclear-inspection regime Gates envisions. Every new rule creates a market for verifying that the rule is being followed. In a world where Gates' agenda gains traction, "AI safety" shifts from a public-relations function to a revenue line.
For investors, the second-order takeaway is not to exit AI but to recognize that the regulatory risk premium is currently near zero. A portfolio built for an unregulated AI boom is not the same as one built for an AI boom that must survive political scrutiny. The former is what most of the market owns.
What to Watch, by Time Horizon
Short term (months): regulatory hearings and legislative proposals. Gates' warnings are already being cited in Washington; expect AI-safety bills to multiply ahead of the next political cycle.
Medium term (one to three years): the first serious attempt at a token tax or robot tax in a major economy, and the first "human reserved" designation in a sensitive sector such as caregiving or education.
Long term (five years plus): whether the international-coordination project gets off the ground at all. Gates himself doubts it will; if it does not, the world enters the AI era with national rules and borderless capabilities — the worst of both worlds.
Three scenarios frame the path:
- Base case: no global body, patchwork national regulation, continued AI investment, rising political friction. The warnings accumulate but do not materially slow deployment.
- Upside case: AI delivers the productivity boom Gates also describes — cheaper clean energy, faster drug discovery, personalized education — and the labor market absorbs the shock through shorter hours and new work.
- Downside case: a high-profile AI-enabled bioweapon or cyber catastrophe validates the "billion deaths" framing and triggers a regulatory overcorrection that crashes AI investment and hands the technology race to less scrupulous actors.
Bill Gates is no longer betting that the good will outweigh the bad. He is betting that only coercion — taxes, reservations, inspectors — can tilt the balance. The market, for now, is betting he is wrong. One of them will be proven right, and the asymmetry of the risk means the cost of being wrong is not borne by the person making the bet.
Explore more exclusive insights at nextfin.ai.
