This weekend in an interview for CBS Sunday Morning1 Dario Amodei again repeated his calls for slowing down the development of the very technology that is starting to make his own company considerable amounts of money.2 The interview was a chance for Amodei to explain the solution he proposed in a blog post earlier this month.3 Despite the rather distracting and at times even ridiculous questions of the interviewer, Amodei provided a sober and careful analysis. His solution you may ask? Regulation.
Amodei himself refers to it as third party evaluators, and compares it to a food inspector. This entity, most likely the government, would verify if AI companies are following the best safety practices—whatever those might be. Amodei is not alone in this, other industry leaders such as Sam Altman and Elon Musk agree. As a response to these remarks many people have wondered why precisely the people controlling and developing the largest AI models in the world, are advocating for regulation of these models.
Why so Serious?
Before taking the AI warnings from Californian Tech CEOs too seriously, the very same Dario Amodei called the early OpenAI model GPT 2 too dangerous to release. Remember that ChatGPT launched using the GPT 3.5 model. How anyone could have thought, even at that time, that GPT 2 was in anyway dangerous seems laughable today. So perhaps as an initial reaction, it might not be too bad to merely laugh?
Of course, once the initial derision has run its course we must nevertheless look carefully at these remarks. Whatever your views on these matters; the importance and influence of the big AI companies, OpenAI, Anthropic, Google, xAI, and Microsoft, are undeniable. Hence, when their CEOs are uniting in the same fearful narrative it would be foolish to not pay attention.
This leaves us with one clear conundrum: why? Why are precisely the CEOs of these companies advocating for their own regulation. It cannot be merely a marketing stunt.4 After all, these comments come months after Amodei warned that Anthropic's frontier models, like Mythos, posed a national-security risk5 to the point where the federal government stepped in.6 The AI companies begged for oversight, and the government gave it to them. Since they continue to shout warnings with the same vim and vigour, this was clearly not an unfavourable outcome. Given the huge amounts of money involved, it seems as unlikely that these CEOs are united solely by a deep moral obligation. Regulation therefore must present large AI companies with some great benefits.
The Fable of AI Scaling
The timing of these AI warnings is of note too. They only started in all seriousness in 2026, at the same time fears of an AI scaling wall started to mount. After the much publicised release of Anthropic's Fable model, and subsequent tumultuous retraction, no new model has outperformed it on the standard benchmarks.7 Improving model accuracy every few months as we saw during the 2024–2025 period, seems over.
Are the warnings of tech leaders to slow down development in that case merely a ploy to distract customers from this reality? Perhaps, but this is certainly not the only reason. It is still possible that these benchmarks are hiding true model improvements on longer more complex tasks.8 The apparent ceiling in accuracy could be an illusion that will be shattered by self-developing AI. Whatever the future of AI scaling may look like, there is a very clear financial incentive today that can explain this perplexing behaviour.
From Regulation to Domination
The financial benefits of the envisioned regulation are quite straightforward. They would protect the incumbent firms, such as Anthropic, by restricting entry into the market.9 Smaller or new companies would struggle to meet the increased overhead, rules, and delays imposed by regulation. The economic incentives for companies that are ahead in the AI race seem to be obvious. Especially when fears on an AI development slow-down are actually rising. If improvements are indeed stagnating, tech giants have even more reasons to prevent competition by other means. Ironically, while many are proposing to slow down development through regulation, they may get their slow down without any intervention.
There is also a simple political reality at play here. While the current Trump administration seems convinced in not regulating AI in any new ways, future administrations are unlikely to stay this course. Tech CEOs may safely assume that a democrat or even a less regulation-shy republican may take over in due time and still impose severe and hindering regulations. A route already paved by the European Union.10 If regulation is inevitable, it may be worthwhile to stay closely involved. Someone will need to define the industry's standards for safety, better it be your own company. Keep a firm hand on the wheel, you might say.
In essence, the AI companies are hoping to get the government to impose an oligopoly by regulation. Forming a cartel would allow them to grow to even more enormous sizes. At the same time, this domination would make it easier to argue that they are indispensable, and "too big to fail". If their monumental capital expenditure on data centres would implode, as many fear, it would be easier for them to convince the government to step in and bail them out from their costly mistakes.
If you are still not convinced, I would point out that I conspicuously left out one company from this story; NVIDIA. Of the big AI firms in the U.S., Jen-Hsun Huang as CEO of NVIDIA, is the only one to dismiss the concerns about the AI development pace.11 Looking at the incentives again gives a very clear picture. A future where there is massive competition with many different companies developing AI models rather than a cartel of tech giants, would be greatly more profitable for hardware provider NVIDIA.
Clearly CEOs of billion dollar AI tech companies are still self-interested, and smart or shrewd enough to act on that self-interest. It is fair to say, that behind the lobbying for regulation, there are some very careful calculations on their part.
From Domination to Calcification
Regulations may hope to avert some Hollywood-style AI doom scenario, but in reality it will only create the incentives that will see the U.S., and therefore the West, fail the AI race. As Thomas Sowell so clearly put it, "Economic policies need to be analyzed according to the incentives they create, and not according to the hopes that inspired them.".12
As the regulations reduce competition, and the tech giants solidify their market positions; the need to innovate will go down precipitously. If you need more proof of this, we can easily compare the situations in regulation-free U.S of Trump with the early regulation-adopting European Union. In the U.S. AI start-ups in places such as San Francisco are growing like mushrooms. Many of these start-ups are developing their own models, such as Thinking Machines Lab, inception, essential AI, Arcee, Liquid, and so on. Across the Atlantic, there are few companies developing similar models. Even the most prominent, Mistral AI, Aleph Alpha, and DEEPL, are continually playing catch-up with U.S. companies.13
When regulation destroys the competitive and vibrant startup landscape in the U.S., the large dominant companies will become the only capable of developing new frontier models. With declining fear of competition, they will become uninterested in pushing the frontier. The costs would no longer justify the rewards. Subsequently, AI development pace would fall. Rather than slow, careful, and safe development as its proponents promise, regulation will give us calcification, slowdowns, and ultimately irrelevance. If you need a vision of how that world looks, you need only look at the European continent.
An Alternative Vision
So what is the alternative? No regulation? A wild west without rules where massive AI companies are entirely unconstrained, and free to use their immense power and money as they choose. Without accountability they could destabilise the global economy, cause massive job loss, manipulate elections, destroy the academia, and other doom scenarios? Of course not. Existing laws already protect against many of the doom scenarios people use to justify regulation, as well as more realistic dangers as argued repeatedly by the U.S Department of Justice.14 And in many cases it is by no means clear that regulating the development process would even prevent the envisioned dangers. After all, the threats posed by AI lie in the use of the technology, rather than the development process.
In the end, we can only guess at the real motivation of tech leaders calling for their own regulation15, but what is clear is the possible cartel forming it could enable. Without regulations on the levels of the pharma or food industries (to name a few), companies would be able to compete more easily with the large tech firms. In turn this competition would diminish much of the power of the big firms, since much of the feared coercion would only be possible through a cartel.
So for all those jumping for the government to take the fight to the AI companies, like Bernie Sanders with his "Ban AI bill", be careful what you wish for. You might make yourself hostage to enormous tech companies that are too big to fail. And for the liberals in the room; remember that one of those companies is run by Elon Musk.