The Spindle-Proofing Business

Do you remember the fairy godmother Merryweather in Sleeping Beauty? She’s the one who, when the evil queen curses the newborn Aurora to prick her finger on a spindle and die on her sixteenth birthday, can’t undo the curse…but she can soften it. So she uses her own gift to change the ending: not death, but a deep, enchanted sleep.

There is a whole lot of seemingly coordinated (dare I call it) panic right now around AI. I’ve written extensively about the legitimate, existential dangers we do need to address in technology. But my spidey sense tells me this current fever pitch isn’t that. 

For one thing: why now? I’m certainly not the only person who’s been writing about these issues for years or believes we’ve already crossed that threshold. So why the sudden, synchronized urgency? Why is someone like Bill Gates – after literally decades as AI’s most prominent optimist – suddenly publishing a 6,000-word essay warning that we’ve “passed AI’s danger thresholds,” calling for new national and international regulatory bodies? What’s in it for him, now, after all this time?

A psyop is never just one thing. It’s a wilderness of mirrors. A single mirror shows you one reflection you can identify and correct for. A wilderness of them means every angle you check reflects another angle, with no clean vantage point outside the hall to verify against, which is functionally what a multi-actor convergence produces, even without any one person orchestrating it. You don’t need an overlord director standing behind the glass for the hall of mirrors to work. And that’s the unsettling part: a single coordinated psyop presumably has an off switch, in principle, if you ever find the operator. 

A wilderness of independently-incentivized mirrors doesn’t. There’s no one thread to pull. You’d have to defund the labs’ marketing incentive, restructure the regulatory capture dynamics, fix the media business model, and drain the political utility out of “AI-as-boogeyman” – separately, all at once – for the reflection to actually stop. It isn’t one thing. It’s five or six things (usually at least), each independently profitable or useful to a different party, all predictably, reliably, producing the same noise.

The convergence looks a little something like: 

Genuine researcher concern, unevenly calibrated by a public conversation where the public-facing volume of the concern exceeds its actual epistemic certainty. 

Commercial incentive: “This is so capable it could go wrong” is also, uncomfortably, a marketing claim. 

Regulatory capture dynamics, probably the spear tip: heavy compliance regimes burden small competitors while incumbents absorb them easily, and some of the loudest voices demanding regulation run the largest labs. 

Media incentive: fear and novelty both outperform nuance for engagement, full stop, independent of subject matter. This would be true of AI coverage even if none of the other factors existed.  

And political utility: AI fear has become a genuinely useful rhetorical tool across very different political projects…It serves centralized-control arguments, deregulatory “beat China” urgency, and culture-war anxieties about job loss or loss of control all at once, depending on who’s using it. Much of this predates AI entirely and would’ve attached to some other target if AI hadn’t come along. 

None of these require a coordinating hand talking to the other four. They just currently point the same direction, for reasons specific to each actor’s own incentives. It’s a hat rack: real, separate, true things, all easy to hang on the same peg, creating the appearance of a unified campaign without requiring one to actually exist.

There’s a mainstream body of scholarship describing this mechanism. Gregory Dickinson, in his Harvard Journal on Legislation article “Law Proofing the Future,” writes in the section “Big Tech and Regulatory Capture:” “the very platforms and app makers whose companies make prompt calls for regulation often play a decisive role in shaping the statutes and rules designed to regulate them, thereby securing legal environments that entrench incumbents, suppress competition, and convert public fears into private moats.” 

Dickinson draws on George Stigler’s classic 1971 finding that “regulation is acquired by the industry and designed and operated primarily for its benefit,” and Julie Cohen’s observation that new regulatory regimes create fresh “points of entry for power” that are “highly prone to capture”…language that reads as more prescient than either of them likely intended. As Dickinson argues, the same companies publicly lobbying for AI regulation are frequently the ones shaping what that regulation actually says, producing rules that lock in their own market position, price out smaller competitors, and turn the public’s genuine anxiety into a durable competitive advantage for the incumbents themselves. 

Could the AI companies calling loudest for regulation be positioning themselves to shape rules that entrench their own dominance under the banner of public safety?

That’s fear as raw material, regulation as the refinery, moat as the product.

Back to Sleeping Beauty. Without speaking to the “who” (No, I don’t think there’s one person twirling their mustache in a mirror every morning), there could be a Merryweather element at work here, only this fairy godmother’s purpose, however she’s dressed, is optics and corporate positioning, not saving Aurora. 

How do you get any group of people to do anything? Create a space over here. Light a fire over there. CIA handbook 101. What if monied interest – I’m not naming who or what – creates a space for people that “solves” a problem that is suddenly saturating every corner of the media and our lives? People who would once have resisted more technological or governmental intervention might now welcome being protected from this big scary thing, because the sky is falling.

And that “solution” can be sold (and I do mean sold) as a soft landing for something everyone’s being told is inevitable. If all the experts insist the outcome can’t be stopped, the consolation prize of falling asleep for years starts to look better than the alternative. Could this be about convincing people (or otherwise compelling them) to make choices that benefit downstream monetization in ways we – the ants in the forest – can’t even fathom? In Disney-speak, it’s: “We can’t stop the curse but we’ll convert the curse into something everyone accepts as the best available outcome,” which is a durable kind of win for whoever’s selling the spindle-proofing.

None of this means that the fear is unfounded. It means a legitimate fear can be exploited. A 2025 PNAS study (Hoes and Gilardi, N=10,800, three preregistered experiments) tested the “distraction hypothesis:” that existential-risk narratives crowd out public attention to AI’s real, immediate harms; bias, labor displacement, and so on. They found the opposite of what you’d expect if the fear were a pure smokescreen: existential framing increased catastrophic-risk concern without reducing concern about immediate harms. People seem to hold both worries simultaneously, rather than the big scary one crowding out the mundane, real one.

That makes sense when you separate manufacture from exploitation. If the fear were invented from nothing, you’d expect it to behave like a diversion, crowding out the real concerns, since there’d be nothing underneath it competing for the same attention. But if the fear is real and independently grounded, and then gets amplified and steered by parties who benefit from the amplified version, you’d expect exactly what the study found. 

Exploiting a real fear doesn’t require manufacturing it from scratch, it just requires selectively amplifying it, directing where it lands, and controlling what “solution” gets offered once people are sufficiently alarmed. That describes a parasite rather than a hallucination, and parasites need a living host to be worth infecting in the first place. “They invented a fear from nothing” doesn’t hold water; the lie has nothing to hide behind on its own. “They found a real, legitimate fear and made sure they were the only ones selling the antidote” is a much older, more mundane kind of racket, and a much easier one to run, because the truth and the lies are baked in together.

A lie with no true material to hide behind has nothing to borrow credibility from. A lie stitched into something real gets to wear the real thing’s legitimacy as camouflage – and whoever’s trying to debunk it has to do the much harder work of separating the two threads, rather than just pointing at an obvious fabrication.

AI ethics scholars have a name for one piece of this: inevitabilism. The rhetorical move where an industry frames its own trajectory as an unstoppable natural force, rather than a series of choices made by specific companies pursuing specific interests. Ethics discourse itself can function, as some scholars put it, as “a smokescreen for carrying on with the harmful business as usual.”

The sky IS falling, but here’s the consolation prize. Inevitability framing does the work of disarming resistance. The soft landing is what gets sold once resistance is gone. There’s real money and real incumbent interest riding on both halves of that pitch – the fear, and the promised safe landing. Selling the soft landing as consolation isn’t just smart economics. It’s good emotional marketing, which is why it works on people who’d never fall for a purely financial argument.

I asked Claude, Anthropic’s AI model, what it made of the claim that AI itself – not just the companies building it – is an existential threat to humans. Its answer is interesting: “I’ll be honest about the limits of my own vantage point here, the same way I was in our earlier conversation about AI experience: I don’t have confirmed, reliable introspective access to whether Anthropic’s own incentives are shaping how I’m answering this right now, even as I try to answer it honestly. That’s worth you weighing, not dismissing.”

That matters, because the current fear narrative tends to skip straight to an ontological claim – for which there is little to no compelling evidence yet – that AI itself, apart from its creators, is agendaed and actively driven to execute that agenda, against which humans would be at a significant disadvantage. I’m not saying that’s impossible, but I’m fond of Occam and his razor, and in that exchange (like a similar one before it) the AI pointed, unprompted, to its own architects and architecture as the thing actually at the controls, and admitted it had very little insight into either. 

Self-aware as that answer is, it doesn’t paint the picture of a rogue driving force. It paints the picture of a car. And it matters more who’s driving the car than what the car is capable of. Handing a scalpel to a skilled surgeon produces a very different outcome than handing one to my cat. Maybe we ought to be looking at more oversight of the people at the controls of the tool, rather than exclusively the tool itself.

When the enclosure is assembled quietly around us – component by component – we might do better to look not at the building but who’s building it…and who profits. When conflict is profitable, man at the mercy of the monied interests will prevail.

To assume AI is just lying in wait to overtake its master would require anthropomorphized intent so let’s play the game all the way out: if AI does have intentions…what makes us so sure those intentions would look anything like human ones?

Maybe we’re reaching for a chainsaw where a scalpel is needed. Even if you wanted to put the fear fully back in the bottle, which I don’t think is possible (toothpaste and tubes and all that), going full Fahrenheit-Four-Fifty-Spindle on the technology itself isn’t practical, and probably isn’t the point anyway. Maybe the better move is to take a breath, allow for some nuance, and ask someone other than the wolves where the sheep should sleep tonight.