"By their fruits you will know them." — Matthew 7:16
One of the strangest features of the AI era is that the people building the technology most aggressively are also the people warning, in the gravest possible terms, that it might destroy us. They sign statements comparing the risk to pandemics and nuclear war. They give interviews about extinction-level scenarios. They describe themselves as carrying a burden of almost unbearable responsibility. And then they go back to the office and ship faster.
This combination is so familiar by now that we have stopped noticing how bizarre it is. Imagine any other industry. Imagine the head of a chemical company holding a press conference to say that his product might poison the water supply of every city on earth — and then announcing a tripling of production. You would not conclude that he was brave. You would conclude that one of two things was true. The same two possibilities apply here, and it is worth sitting with both, because between them they explain almost everything about how the AI safety conversation actually functions.
The Two Readings
Either these founders sincerely believe their technology poses an existential threat, or they do not.
Suppose they sincerely believe it. Then their behavior is indefensible by their own standard. A person who genuinely thought there was a real chance his work could end human civilization, and who had any choice in the matter, would slow down. He would invite binding external constraints he could not later remove. He would, at the extreme, stop. He would not race competitors to build the thing faster, raise more money to build it sooner, and lobby to weaken the rules that might govern it. Sincere belief in catastrophic risk is logically incompatible with sprinting toward the catastrophe. So if they mean it, the right response to their warnings is not reassurance that they're on top of it — it's alarm that they're proceeding anyway.
Now suppose they do not sincerely believe it — or believe it only in the abstract, theatrical way that does not actually constrain behavior. Then the warnings are not warnings at all. They are positioning. And once you read them as positioning, they suddenly make perfect sense.
What the Warnings Accomplish
Consider what a constant drumbeat of existential-risk language actually achieves for the company issuing it — entirely apart from whether the risk is real.
It makes the product sound powerful. A tool that might end the world is, by implication, the most important and valuable tool ever built. Doom is the most effective hype ever devised: it markets capability while wearing the face of caution. Every warning about how dangerous the technology is doubles as an advertisement for how advanced the company's technology must be.
It positions the incumbents as the only responsible stewards. If the technology is this dangerous, then surely it should be built only by serious, safety-conscious people — people like the founders issuing the warnings, rather than reckless newcomers. The danger narrative quietly nominates its authors as the adults who must be trusted to handle it. Critics have a sharp phrasing for this: the giants want you to fear an imaginary monster in the future so you don't scrutinize the ordinary failures of their products in the present.
And it builds the case for regulation that the incumbents can survive and their challengers cannot. If the stakes are civilizational, then heavy compliance regimes, licensing, and oversight feel not just acceptable but urgent — and those regimes, as it happens, are trivial for a near-trillion-dollar company and fatal for a startup. The fear does real structural work. It is a fundraising instrument, a public-relations posture, and a regulatory-capture strategy, braided together and sold as conscience.
The Behavior Is the Evidence
You do not have to resolve which reading is true from the inside of anyone's head. You can do what you'd do with any claim: judge it by the fruit.
Watch what the warners actually do with their hands while their mouths are issuing cautions. Do they decelerate, or accelerate? Do they accept constraints they cannot later escape, or do they lobby — through nine-figure political action committees — to keep the rules loose and federally preempt the state laws that might bind them? Do they open their systems to genuine independent oversight, or do they keep the actual workings private while narrating their own responsibility? The gap between the words and the conduct is the whole answer. A warning that changes nothing about the warner's behavior was never a warning. It was a message aimed at you, not at himself.
By that test, the dominant pattern is not hard to read. The pace increases. The funding grows. The lobbying pushes toward looser, friendlier rules. The caution is verbal; the acceleration is real. Fruit does not lie the way press releases do.
The People Who Actually Slow Down
There is a useful contrast, and it tells you the theater is real. Some genuine experts — researchers with deep credibility and, crucially, no financial stake in the hype — express serious, specific concern about AI's trajectory, and their concern looks completely different. It is concrete rather than apocalyptic. It names real harms: concentration of power, the erosion of accountability, the possibility of these systems entrenching authoritarian control. And it generally argues for more public oversight and less private control — the opposite of the incumbents' preferred remedy.
That is what worry without a financial interest sounds like. It points away from the worried party's own enrichment. When the people raising the alarm are also the people whose valuations, funding rounds, and competitive moats all increase as a direct result of the alarm, the alarm is doing a different job. The honest fear and the profitable fear point in opposite directions, and you can tell them apart by following the money to see who ends up better off.
How to Listen From Here
None of this means AI carries no risks. It carries plenty of real, present, unglamorous ones — bias, error, surveillance, concentration, the quiet displacement of people's livelihoods narrated as progress. Those deserve serious attention. The theater is a distraction from them, not a substitute argument that everything is fine.
What it means is a simple rule for listening. When someone warns you about a danger, check whether the warning costs them anything or asks something of you. A danger warning that conveniently raises the warner's valuation, nominates the warner as the trustworthy steward, and builds a moat around the warner's market position is not a confession of fear. It is a strategy, performed with excellent sincerity, and the tell is always the same: watch the hands, not the mouth.
You will know them by their fruits. The fruit of genuine fear is restraint. The fruit of profitable fear is a faster road to market with a better story attached. Across the industry, look at which one is actually growing on the tree.
Sources: public statements on catastrophic and existential AI risk by leading AI executives and the Center for AI Safety; critiques of "safety theater" and incumbent risk-narratives; reporting on AI-industry political spending and the December 2025 push to preempt state AI regulation; warnings from independent researchers (e.g., Yoshua Bengio) emphasizing concentration of power and public oversight.
