By Will Hyland — Founder building for the AI era
A warning about artificial intelligence can also be a bid for power.
When someone tells us a technology could threaten civilization, we should examine the danger carefully. We should also examine what they want us to do about it. Who would control the technology? Who would write the rules? Who would be permitted to compete? Who would collect the revenue?
Those questions become especially important when the people sounding the alarm have financial, professional, or political interests in the proposed response.
AI may present serious risks. Fear of those risks may also help concentrate its benefits. Both possibilities deserve scrutiny.
In a September 2026 interview with NBC’s Meet the Press, Bill Gates said AI was “powerful enough to drive events that, you know, cause a billion deaths.” He was discussing malicious people using powerful AI tools. He was describing catastrophic potential, rather than predicting a specific death toll. Bloomberg’s account of the interview (opens in a new tab)
That distinction matters. So does the effect of the language.
A billion deaths is a number large enough to overwhelm almost any competing concern. Once it becomes the frame for a discussion, questions about competition, affordability, privacy, and access can seem secondary.
Yet those questions help determine who will hold power in an economy increasingly shaped by AI.
We should be able to investigate catastrophic risks without handing the people describing them a blank check.
Consider the difference between identifying a danger and establishing the right to govern it.
A company may have valuable knowledge about the systems it develops. That knowledge deserves attention. But technical expertise does not automatically make the company an impartial judge of which competitors should be allowed to operate, what independent researchers should be permitted to inspect, or which infrastructure everyone else should have to use.
The commercial interest remains, even when the safety concern is sincere.
We do not need to assume secret coordination to recognize this. Organizations can independently favor policies that protect their position. Political leaders can favor arrangements that expand their authority. Experts can become more influential when the problem they study becomes a national priority.
An outcome can serve powerful interests without everyone involved sharing a hidden plan.
The strongest version of the argument is therefore about incentives and institutional design. Private motives are harder to establish. Proposed rules, business relationships, and barriers to entry can be examined directly.
Microsoft’s May 2023 AI governance blueprint provides a concrete example. It proposed licensing for highly capable foundation models and licensed datacenters for certain systems controlling critical infrastructure. These were specific proposals connecting safety oversight with permission to develop and deploy powerful technology. Microsoft’s governance blueprint (opens in a new tab)
There may be circumstances in which such requirements are justified. A system with access to critical infrastructure warrants different scrutiny from a tool that helps someone organize meeting notes.
But licensing also raises an unavoidable economic question: who can afford to qualify?
Imagine a compliance regime requiring substantial legal work, specialized security teams, expensive evaluations, and lengthy approval processes. A large company might absorb those costs across millions of customers. A new entrant might never get to its first customer.
The same rule can therefore reduce a risk and reduce competition.
That possibility does not settle the policy question. It makes the competitive effects part of the policy question. Advocates should have to explain why a proposed restriction is necessary, how it reduces the stated danger, and whether a less restrictive alternative could accomplish the same objective.
The existing market structure makes this especially important.
In January 2025, the Federal Trade Commission published findings on partnerships between major cloud providers and AI developers. It identified potential competition concerns involving access to computing resources, switching costs, and access to sensitive business information. The report did not establish that fear created those arrangements. It showed why additional restrictions would enter a market already facing questions about dependence and concentration. FTC findings (opens in a new tab)
A safety regime should not quietly turn an existing commercial advantage into a permanent condition of entry.
There is another possible benefit to dramatic warnings: they can elevate the people delivering them.
A technology executive who warns about civilization’s survival is speaking on a larger stage than an executive explaining a product. The warning can create a claim to public leadership. It can make that person a sought-after adviser on questions extending far beyond their technical expertise.
For politicians, the equivalent opportunity is to become the protector: the person empowered to authorize, prohibit, monitor, or fund.
These are plausible incentives, not proof that any particular warning is dishonest. But they explain why financial gain is too narrow a lens. Professional relevance, access, prestige, and authority can matter as well.
The public should be particularly attentive when a warning arrives with the implication that only a small circle of insiders can understand the threat well enough to manage it. Expertise should help people examine a claim. It should not require them to accept the claimant’s authority indefinitely.
Fear can also work in the opposite direction.
One argument says AI is so dangerous that development must be restricted. Another says falling behind in AI is so dangerous that development must be accelerated.
The White House’s July 2025 AI Action Plan connected winning the AI race with preserving American economic and military leadership. Its agenda included reducing regulatory barriers and expanding infrastructure. White House announcement (opens in a new tab)
Those priorities may have legitimate national-security foundations. They also illustrate how a threat narrative can support faster development and favorable conditions for industry.
This creates a possibility worth watching: restrictions that make entry harder for outsiders, alongside government support that makes expansion easier for established players.
Whether that happens must be demonstrated through actual policies. The important point is that fear does not have a single political direction. It can justify both restraint and acceleration, depending on whose activity is being discussed.
That is also why treating the technology industry as one unified bloc weakens the analysis.
Different companies have different advantages. Their preferred rules can reflect those differences.
In July 2024, Mark Zuckerberg argued that open AI development could improve safety and reduce dependence on closed providers. That position also fit Meta’s competitive interests. It deserves evaluation on its merits, with the same attention to incentives that we apply to arguments for tighter restrictions. Zuckerberg’s statement (opens in a new tab)
Commercial alignment does not automatically invalidate an argument. Neither does a public-interest vocabulary establish its independence.
The same fairness should apply to Gates.
In his August 2026 essay, Gates acknowledged continuing financial ties to technology and said investment profits would support his foundation. He also advocated taxes on AI tokens and robots, protecting some human roles, and involving workers and communities in decisions about the transition. Those positions complicate the claim that his agenda is simply to maximize technology-industry profits. Gates’s essay (opens in a new tab)
We can question the evidence behind his warnings and the consequences of his proposals. The quotation alone does not establish that he is manufacturing fear for personal enrichment.
A credible critique needs that discipline. Otherwise, it becomes another emotionally powerful story asking readers to accept more than the evidence supports.
The strongest objection to this entire argument is straightforward: what if the danger is severe, and powerful people are warning us because they understand it?
That possibility must remain open.
The February 2026 International AI Safety Report described substantial expert disagreement about scenarios in which humans lose control of AI. It distinguished emerging capabilities relevant to those scenarios from the capabilities needed to produce such an outcome. It also emphasized the uncertainty surrounding their likelihood and timing. International AI Safety Report (opens in a new tab)
Uncertainty does not mean safety. Waiting for a catastrophe to prove that a precaution was needed would be a terrible standard.
But uncertainty also does not validate every frightening scenario or every proposed restriction. A responsible response must connect the evidence, the possible harm, and the intervention.
Malicious use, unreliable automation, and autonomous loss of control are different problems. A measure that addresses one may do little about another. Combining them into a single image of impending disaster makes the public debate less useful.
I want AI claims tied to observable performance.
What can the system do? Under what conditions? How reliably? With what access to tools or infrastructure? What happens when someone tries to misuse it? Which safeguards work, and where do they fail?
An impressive demonstration does not establish dependable capability. An alarming demonstration does not, by itself, establish the probability of catastrophe.
Both deserve independent evaluation.
This matters to people building outside the largest institutions. If AI can lower the cost of research, software development, administration, and other useful work, more individuals and small businesses may be able to participate in activities previously beyond their resources.
That opportunity is conditional. Access, affordability, skills, reliability, and customer demand all matter.
A public conversation dominated by catastrophe could discourage some people from learning and experimenting while better-resourced organizations continue adopting the technology. That is a hypothesis worth testing, rather than an established explanation for anyone’s rhetoric.
It is also a reason to communicate risk precisely. People need enough understanding to make practical decisions. Fear that leaves them feeling helpless serves them poorly.
I bring an interest to this debate myself: I am building AI ventures. I want room for independent builders to compete and for individuals to benefit. That preference should be visible, and it does not exempt the systems I develop from scrutiny.
The standard I would apply is simple.
Serious warnings should come with evidence proportionate to the claim. When sensitive details cannot safely be published, qualified independent reviewers should still be able to examine them.
Proposed controls should be tied to specific risks and evaluated for their effects on competition, privacy, and access. Where precaution is necessary before the evidence is conclusive, the uncertainty and the conditions for revising the policy should be explicit.
People advocating restrictions should accept meaningful scrutiny of their own systems. People advocating openness should explain how they would handle capabilities whose release could create serious harm.
And rules should be revisited as evidence changes. A temporary precaution should not become a permanent commercial privilege simply because its beneficiaries prefer it that way.
The question of who benefits from AI fear belongs inside the safety debate. Public protection and concentrated power can emerge from the same policy. We have to examine both.
Whenever an influential person makes a frightening AI claim, follow it through to the proposed response: the permissions it creates, the costs it imposes, the information it makes available, and the authority it assigns.
Then ask whether that response makes people safer while preserving their ability to understand, use, challenge, and benefit from the technology.
The people who define AI’s dangers will help shape AI’s future. Their claims should earn our attention. Their proposed powers should earn our consent.