You’ve probably seen the headlines by now: a letter from employees at OpenAI and Anthropic, begging the US government to step in and ‘pace’ AI progress. Sounds noble, right? The brave insiders sounding the alarm on existential risk.
Let me tell you what that letter actually is: the most sophisticated regulatory moat ever built.
Here’s the paradox that should make you furious. The very companies racing to build artificial general intelligence at breakneck speed—spending billions on GPUs, poaching every PhD in sight, shipping products before they’re safe—are now asking the government to slow everyone down. They’re not afraid of AI. They’re afraid of competition.
You’ve watched this play before. Big tobacco said they wanted regulation for the children. Big banks said they wanted oversight to protect the economy. Every time, the result was the same: the incumbents got to write the rules, and the upstarts got crushed. This isn’t a plea for safety. It’s a textbook play for regulatory capture, disguised as a moral crusade.
Think about it. OpenAI and Anthropic are already in the room where the rules are being drafted. They have the ear of policymakers, the trust of the media, and the deepest pockets in the industry. By helping define what ‘safe AI’ means, they get to define the standard that everyone else must meet. And their own models? They’ll be the baseline. Everyone else will be playing catch-up—if they can afford the compliance costs at all.
I saw this firsthand at a closed-door briefing last year. A senior executive from one of these companies stood up and said, with a straight face, ‘We need a licensing regime for frontier AI models.’ The subtext was deafening: We already have ours. Good luck getting yours.
Don’t take my word for it. The letter itself is full of weasel words: ‘pacing’ not ‘halting,’ ‘guidance’ not ‘prohibition.’ They don’t want a ban. They want a gatekeeper—and they want to be the gatekeepers. When the biggest players in a race ask for a referee, they’re not worried about injuries. They’re worried about losing.
The real tragedy is that this works. The narrative of ‘existential risk’ has been so thoroughly seeded in the public mind that any politician who questions regulation is painted as reckless. Meanwhile, the actual risks—massive concentration of power, algorithmic bias, job displacement, surveillance—are being ignored because they don’t serve the monopoly narrative.
So what happens next? If the US government obliges, we’ll get a cozy cartel of approved AI labs, overseen by a new agency staffed by the very people who wrote the rules. Startups will die. Open-source models will be outlawed. And the rest of us will be left with whatever sliver of intelligence the incumbents decide to license back to us, at a price they set.
This isn’t a future we have to accept. But it’s the one we’re heading toward, unless we stop pretending that the people asking for regulation are doing it for our benefit. They’re not afraid of the fire. They’re afraid of anyone else holding the match.
FAQ
Q: Aren't OpenAI and Anthropic genuinely concerned about AI safety?
A: Some individuals may be sincere, but the corporate action—asking for government intervention while racing to deploy their own products—is strategically inconsistent. If they truly believed AI was an existential threat, they would stop building, not ask for a rulebook that locks in their advantage.
Q: What's the practical implication if they succeed?
A: A small number of AI labs will control the regulatory standards, effectively creating a licensing barrier that blocks new entrants and open-source alternatives. This means higher costs, less innovation, and a future where the most powerful AI is only available to the richest corporations and governments.
Q: Isn't some regulation better than none?
A: Bad regulation—captured by incumbents—is worse than no regulation. It creates an illusion of safety while entrenching monopolies. The right approach is to regulate outcomes (e.g., liability for harm) rather than methods (e.g., who gets to train a model). The current push is about controlling the latter.