Hey everyone,
What if the danger is real — and the story being told about the danger is still a psyop?
Something strange is happening at the frontier of artificial intelligence.
The companies racing hardest to build the most powerful systems in human history have begun warning us—with remarkable consistency—that those systems may be too dangerous for anyone else to build.
The warnings are getting louder. The timelines are getting shorter. The language is moving beyond the familiar concerns about bias, hallucinations, copyright, job displacement, or people using chatbots to cheat on homework. We are now being told to prepare for systems that can escape, deceive, blackmail, hack, replicate, design weapons, seize control of the internet, and perhaps end human civilization altogether. The people delivering this news are not critics standing outside the machine. They are the people building it, selling access to it, raising billions of dollars against it, and racing one another to make the next version even more powerful.
And now, having created the danger, they would also like to define it for us.
This is where the conversation gets slippery, because there are two things that can be true at once. Advanced AI may carry serious risks. Some of those risks may be catastrophic. A technology does not have to be conscious or evil to become dangerous at scale; it only has to be capable, widely deployed, and connected to systems that matter. We should absolutely ask hard questions about what these models can do, who can direct them, and what happens when their capabilities exceed our ability to supervise them.
But the danger may be real and the story being told about the danger can still be a psyop.
I don’t mean a psyop in the cartoon sense: a handful of executives meeting in a dark room to invent a fake apocalypse. They don’t need to. Incentives can coordinate a narrative all by themselves. Every frontier company benefits from the public accepting the same basic premise: *what we have built is so powerful, so complicated, and so potentially dangerous that only institutions like ours can be trusted to control it.*
That’s a hell of a product position.
Think about what the fear accomplishes. It tells investors that these companies aren’t selling another software tool; they’re building the most consequential technology in history. It turns embarrassing model behavior into evidence of astonishing capability. It makes governments dependent on the labs for technical interpretation. It makes the labs’ own evaluation systems the natural basis for regulation. It allows enormous infrastructure costs, closed models, restricted access, and centralized control to be presented not as commercial choices, but as acts of public responsibility.
Best of all, it turns competition into recklessness.
If a smaller company releases an open model, it isn’t merely competing. It is endangering humanity. If an independent researcher questions the frontier labs’ safety claims, she isn’t exercising skepticism. She is failing to take the threat seriously. If regulators hesitate to build laws around the labs’ preferred frameworks, they aren’t protecting competition. They are gambling with civilization.
The frame is almost impossible to argue with once you accept it.
Fear compresses judgment.
That may be the most important part of the operation. Under ordinary circumstances, we would ask a long list of inconvenient questions. What exactly did the model do? Under what conditions did it do it? How many attempts were required to produce the frightening result? Were its safeguards deliberately removed? Was it operating in the world or inside a scenario designed by the company testing it? Can the result be reproduced independently? What evidence supports the timeline? Who benefits if this particular interpretation becomes conventional wisdom?
Those aren’t dismissive questions. They are the questions responsible people ask before granting a private company extraordinary authority. But attach the phrase “human extinction” to the discussion and suddenly asking for evidence sounds unserious. The person demanding clarity becomes the dangerous one.
Meanwhile, the proposed solution has a curious tendency to resemble the companies proposing it. Safety requires immense compute. It requires proprietary evaluations. It requires permanent monitoring, specialized teams, government partnerships, licensing regimes, and compliance programs that only the largest companies can afford. The frontier labs already possess all of those things. A regulatory system built around them would not merely make AI safer.
It would make today’s frontier permanent.
That doesn’t mean every safety proposal is corrupt or every person working on alignment is lying. Plenty of serious people inside these companies are sincerely alarmed. I suspect many of them should be. But sincerity does not cancel incentive, and concern does not confer sovereignty. A company can believe its warnings and still benefit enormously from the world organizing itself around them.
The fire department can be genuinely worried about the fire while quietly neglecting to mention that it owns the hydrants, wrote the building code, and started the fire in the first place.
What makes this moment especially strange is that the companies asking everyone else to slow down do not appear particularly interested in surrendering their own position. They cannot stop because a competitor might continue. They cannot disclose everything because the information could be dangerous. They cannot allow unrestricted access because ordinary people might misuse it. They cannot accept regulation that allows foreign rivals to gain ground. Somehow every road through the existential crisis leads to the same destination: the frontier companies keep building, keep deciding, and keep owning the result.
We are invited to confuse their inability to stop competing with a mandate to rule.
Maybe the systems are as powerful as they say. Maybe they will become more powerful than most people can currently imagine. If that’s true, we need more judgment, more independent evidence, more distributed scrutiny, and more people capable of challenging the story—not a priesthood whose authority rests on exclusive access to the god it created.
We should regulate harmful conduct. We should demand disclosure when frontier systems cross meaningful capability thresholds. We should require evidence, audit the auditors, and build consequences for companies that recklessly connect unreliable systems to critical infrastructure. We should take security, provenance, consent, labor disruption, market concentration, and human agency seriously. There is no virtue in being cavalier simply because powerful people have learned to weaponize caution.
But we should also notice when safety becomes a moat.
We should notice when hypothetical future catastrophe consumes all the oxygen that might otherwise be used to discuss present ownership. We should notice when the builders of closed systems describe openness itself as the threat. We should notice when the required cure happens to eliminate smaller competitors, independent builders, and public alternatives. We should notice when companies ask to be regulated according to standards they are uniquely positioned to satisfy.
Most of all, we should resist the demand that fear replace judgment.
The frontier companies may be right that AI is dangerous. They may even be understating some of the risks. But they are not neutral observers, and they have not earned the right to translate capability into authority without challenge. Building the most powerful machine does not make you the rightful governor of everyone who might use one.
Perhaps AI really could become a monster.
If so, the last people entitled to define the monster alone are the people racing to own it.
Building the most powerful machine does not make you the rightful governor of everyone who might use one.
Talk soon,
Todd