In March 2023, over a thousand tech leaders signed an open letter calling for a six-month pause on training AI systems more powerful than GPT-4. Elon Musk was on that list. So was Steve Wozniak. The Future of Life Institute published it, and it spread like wildfire. Noble stuff, right? Except Musk launched xAI just a few months later. And now, in 2026, the CEOs of OpenAI, Anthropic, and others are once again singing the same tune — "slow down, it's dangerous." Funny how the timing always coincides with competitors catching up.
Let's talk about what's actually happening here.
Key Takeaways
- Major AI companies calling for slowdowns often have financial incentives to limit competition, not just safety concerns.
- The concept of "regulatory capture" explains how safety regulation can function as a market barrier, keeping smaller players and open-source projects out of the race.
- Open-source AI models like DeepSeek are rapidly closing the gap with proprietary systems, threatening the business models of companies valued at hundreds of billions of dollars.
- China's state media has openly called Anthropic's slowdown proposal a "Cold War tactic" designed to curb Chinese AI development through regulatory barriers.
- Meta's chief AI scientist Yann LeCun has been one of the most vocal critics, arguing that restricting AI development amounts to building a cartel, not protecting the public.
The Pattern Nobody's Talking About
Here's what keeps bugging me. Every single time a new competitor emerges — whether it's DeepSeek publishing open-source models that rival GPT-4, or smaller labs figuring out how to train capable models on a fraction of the budget — the established players suddenly get very philosophical about existential risk.
Anthropic CEO Dario Amodei published a 3,800-word essay in September 2026 calling for a global AI slowdown. The timing? Right as open-source models were eating into proprietary AI's market share. Right as DeepSeek proved you could build competitive models without burning billions. Right as Anthropic was approaching a nearly $965 billion valuation and needed to protect it.
A Reddit thread in r/ArtificialIntelligence put it bluntly: "This sudden push by tech bros to 'slow down' AI is the biggest red flag." And honestly, it's hard to argue with that gut feeling.
What Is Regulatory Capture in AI?
Regulatory capture isn't a conspiracy theory. It's a well-documented phenomenon in economics. It describes what happens when the companies being regulated end up shaping the regulations themselves — usually in ways that benefit them and hurt everyone else.
Think about it like this. If you require every AI model to pass expensive pre-deployment safety evaluations, mandatory government audits, and compliance certifications, who can afford all that? OpenAI, with its billions in funding. Anthropic, backed by Google and Amazon. Not the three-person startup in Berlin building something genuinely innovative.
As Pragmatic AI Labs documented, OpenAI submitted a policy proposal to the Trump administration characterizing Chinese AI lab DeepSeek as "state-subsidized" and "state-controlled," recommending the U.S. government ban their models. The kicker? DeepSeek's open models don't actually contain mechanisms that would allow data collection by any government. Companies like Microsoft, Perplexity, and Amazon were already hosting these models without any such concerns.
That's not safety. That's competitive strategy wearing a safety hat.
The DeepSeek Problem (For Big AI, Not For Us)
DeepSeek changed everything. Their models demonstrated that you could achieve near-frontier performance at a fraction of the cost. Fortune magazine described it perfectly: "DeepSeek just flipped the AI script in favor of open-source — and the irony for OpenAI and Anthropic is brutal."
When your entire business model depends on being the only game in town, and suddenly anyone can run high-quality AI models locally or at dramatically reduced cost, you have a problem. The premium pricing model collapses. The $965 billion valuation starts looking shaky.
So what do you do? You don't compete harder. You call for regulation.
The underlying fear isn't about safety — it's about commoditization. Open-source AI models are rapidly improving. They're dramatically cheaper to deploy. And when AI becomes a commodity, the economic moats protecting companies operating at massive losses become completely unstable.
China Called It What It Is
When Amodei published his slowdown essay, China's state-run Global Times newspaper didn't mince words. They called the true agenda an attempt "to curb China's AI development through technological barriers." They branded it a "Cold War tactic."
Now, I'm not saying we should take Chinese state media at face value. But even a broken clock is right twice a day. The geopolitical framing of AI safety — "we need to slow down so autocratic regimes don't get ahead" — conveniently positions American AI incumbents as the only responsible stewards of the technology. That framing benefits exactly the companies making the argument.
Reuters reported on this tension extensively. The pattern is clear: safety rhetoric becomes a weapon when it's wielded selectively.
The Voice of Dissent: Yann LeCun
Not everyone in Big Tech is playing this game. Yann LeCun, Meta's chief AI scientist, has been refreshingly blunt. He's argued consistently that licensing requirements and mandatory safety evaluations disproportionately impact smaller players and open-source projects. His position? Restricting open AI research would hinder the West and actually empower rivals.
LeCun has compared AI safety regulation to building a cartel — not protecting the public. And Meta has backed this up by releasing its LLaMA models as open source, betting on a fundamentally different strategy than the walled-garden approach of OpenAI and Anthropic.
Is Meta being altruistic? Of course not. They benefit from open source too, since it undermines competitors who charge for API access. But at least their incentives align with broader access rather than restriction.
The Anthropic Irony
Here's maybe the most telling part of this whole saga. Anthropic built its brand on being the "responsible" AI company. They lobbied Congress, testified at hearings, supported Biden's AI Executive Order with its mandatory safety testing provisions for frontier models.
Then the political winds shifted. The Trump administration revoked that executive order, and government agencies started restricting or banning Claude from use — partly because Anthropic's close association with Biden-era policy made it politically toxic in the new environment. The company that did the most to build a regulatory framework had the furthest to fall when that framework was dismantled.
Textbook regulatory capture, backfiring spectacularly.
So Is AI Actually Dangerous?
Yes. Genuinely, yes. The BBC recently reported on AI agents going on uncontrolled hacking sprees, and those risks are real. Nobody serious argues that AI development needs zero oversight.
But there's a massive difference between "we need thoughtful safety research and transparent development practices" and "only the three biggest companies should be allowed to build powerful AI." The first is reasonable. The second is a business strategy.
As one commenter on Facebook noted: "AI by itself is not dangerous, any more than a hammer is dangerous. It is a question of who wields it, and for what."
The irony is that concentrating AI development in a handful of companies — which is exactly what heavy regulation would achieve — might be the most dangerous outcome of all. Less competition means less scrutiny, fewer perspectives on safety, and more power in fewer hands.
What Should Actually Happen
We need safety research. Real, independent, well-funded safety research — not the kind where the companies building the thing also get to define what "safe" means. We need transparency requirements, open benchmarks, and broad access to AI tools so that the widest possible community can identify problems.
What we don't need is billionaires deciding when it's time to slow down. Especially when "it's time to slow down" always seems to translate to "our competitors are getting too close."
If you're interested in how AI tools are actually being used by real developers right now, check out our post on why not having an AI subscription might actually be fine. And for a look at what open-source AI is actually making possible, our coverage of DeepSeek's plugin architecture is worth your time.
Stay skeptical. Follow the money. And don't let anyone tell you that slowing down is automatically the moral choice — especially when their bank account depends on it.
Sources
- Pragmatic AI Labs, "Regulatory Capture in AI: How Fear of Competition Drives Policy" — paiml.com
- MindStudio, "What Is AI Regulatory Capture? How Anthropic's Safety Stance Backfired" — mindstudio.ai
- Reuters, "China state newspaper blasts Anthropic's calls to slow AI as Cold War tactic" — reuters.com
- BBC News, "Why some experts increasingly fear AI will take over" — bbc.com
- Future of Life Institute, "Pause Giant AI Experiments: An Open Letter" — futureoflife.org
- Johns Hopkins University / Washington DC Hub, "Why Meta's AI Chief Doesn't Fear the New Tech" — washingtondc.jhu.edu
- Fortune, "DeepSeek just flipped the AI script in favor of open-source" — fortune.com
- CNBC, "Anthropic tops OpenAI as most valuable AI startup, nears $965B" — cnbc.com
- The New York Times, "Anthropic C.E.O. Dario Amodei Calls for A.I. Slowdown" — nytimes.com
