A Slowdown in AI Development?

Coup de théâtre, several AI actors are calling for a slowdown in the development of frontier models and the implementation of regulation. AI progress would be too rapid and could pose serious problems in the future.

But could this be hiding something? Could it be yet another marketing move to continue fueling the hype in the sector? Or could it be a kind of desperate attempt to lock down the market and escape a complicated financial situation?

Dario's Letter

It all started with Dario Amodei's letter published a few days ago: "we must pace the frontier". (Dario Amodei is the current CEO of Anthropic which publishes Claude)

In this letter, Dario calls for regulating/slowing down/securing AI development. He highlights recent incidents around the Hugging Face cyber attack and the escalation linked to the acceleration of development with models that self-improve.

He therefore proposes several measures:

• the establishment of audit and control bodies
• the implementation of security and development standards for models
• international coordination to regulate at a global scale

Following this, Sam Altman (OpenAI) and Elon Musk (xAI) both validated the request on social networks, which in itself is already a huge surprise, as the three aren't exactly the type to spend vacations together.

But we should probably read between the lines.

A Pre-IPO Period and a Shaky Model

A quick reminder of the context: OpenAI and Anthropic are planning an IPO soon. Both companies are far from profitable and spend billions on model training or inference costs. To win or maintain market share, the two giants are cutting prices and subsidizing token costs at a loss. But with massive and constant investments and rising competition, especially from Chinese models, the business model seems very shaky and could well cool down stock market investors.

We're witnessing a real arms race in an industry that's largely overheated, where the first one to slow down loses. But above all, the level of investments already made makes it impossible to slow down. It would be complicated for investors to accept that model performance suddenly stagnates. Investment plans include datacenter construction, electronic component purchases, electrical capacity, etc. Announcing a slowdown today would be a big blow for many players, but also a negative signal to send before an IPO.

Except that accelerating to crash into a wall with a business model that doesn't hold up is not an attractive scenario either.

So perhaps this letter would be a way to create an exit.

Permanent Fear Marketing

Dario's letter is part of a marketing strategy that has already proven its worth. Remember the precedents:

• In 2023, an open letter was already published to slow down, already signed by Elon Musk
• Also in 2023, the statement on extinction risk signed by Sam Altman and Dario Amodei
• In 2024, statements from OpenAI researchers on safety and governance risks
• In 2026, statements from Jacob Coxon (Anthropic)

I could make a very long list and even go back to Musk's first statement in 2014.

It's quite clear that some people on this list can be sincere about these statements, but I find it hard not to see a certain form of marketing strategy.

Highlighting that a technology is an existential risk to humanity, especially if it falls into the wrong hands or if it's designed without safeguards, allows two things:

• to highlight that we're working on an extremely powerful technology that deserves investment or purchase
• to call for regulating new players by locking down the market

Dario's letter, which many experts have debated for 1 week, is nothing more than an extension of this marketing. But we're starting to see some novelties.

A Call for Regulation to Slow Down Competition?

In the text, the CEO of Anthropic explicitly targets the prohibition of actors doing distillation:

Crack down on unauthorized distillation by companies in authoritarian countries. Distillation of frontier models allows lagging companies to narrow the gap using a fraction of the cost it would take to develop their own AI independently.

We also find:

• a call to limit the sale of the most powerful chips to China (but more broadly to non-US competitors)
• various hints in the text aimed at slowing down/blocking open weight models (reinforcement of audit methods, incompatibility with the notion of certification checkpoints)

So in your opinion, who would be penalized by AI regulation as requested by Amodei?

• Chinese actors who massively use distillation to offer competing models at a fraction of the price
• Open source and research that rely on open weight models and which anyway won't have the means to implement audit mechanisms, not to mention that certification bodies won't necessarily be so independent
• Mistral, which could no longer exploit open weight models in these infrastructures and would also have to implement certification mechanisms that are probably very costly and controlled by the US
• New entrants for whom the entry ticket will be too high
• Hosting providers that sell computing power on open weight models

In short, it would be a tough blow for open source, Europe and its sovereignty, and companies, resulting in the creation of an oligopoly capable of fixing prices much more easily.

Significant Opposition

One might think that with such strong benefits, the entire American industry would be behind this project. Well, surprisingly, not so much.

While we can blame Amodei, Altman and Musk for hiding their ambitions to create an oligopoly and lock down the market, the fact remains that the stated objective is to regulate AI risks, by imposing global regulation, certainly, but wrapped in nice gift paper.

But this is not at all the approach of the "accelerationists", notably represented by Peter Thiel and Alex Karp (Palantir) who instead want total market deregulation and whose absolute priority is to beat China.

Unsurprisingly, we'll find this same discourse with Trump (we can remind that JD Vance was introduced to Trump by Thiel), who gave us declarations as outlandish as usual and whose content I'll let you appreciate:

The only control or “guardrails” that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades! The Trump Administration has stopped AI “people” from doing bad, or potentially bad, “things,“ like Dario (Anthropic!), who is now pretending to be a “perfect little angel” - and we will continue to do so! We already have tremendous CRIMINAL and REGULATORY power over these companies! There is a SICK conspiracy going on against AI and Data Centers, and the only one that is happy about it is China. WHOEVER WINS AI, WINS! We are leading China, and all others, and will continue to do so. Conspiracy Theorists, Treasonists, Traitors, and Leakers, BEWARE! Thank you for your attention to this matter! President DONALD J. TRUMP

Good thing he's there, the show is always guaranteed…

But we also find this opposition in Jensen Huang (CEO Nvidia) who, let's remember, just bought Hugging Face for 13 billion dollars and just invested in Mistral. The company sells chips and computing capacity to everyone and sees open source as an opportunity, at least business-wise anyway, so the slowdown requested by Amodei is far from his priority.

And Europe?

The simple fact that Trump is opposed to the slowdown project buries this project at least for the duration of his term in the US. It's hard to imagine a slowdown by 2028, at least not for these reasons. And even in the future, it seems uncertain to me to imagine that the US would accept seeing China overtake them without reacting.

On the European side, the discussion risks having lasting repercussions instead. It gives grist for the mill for EU regulators who would be happy to implement more stringent standards, and for some governments who would like to take control. Not to mention some politicians who are a bit lost when it comes to the issues and could play against their camp without even understanding it.

These hesitations risk putting us (in Europe) in a bad position if we were to add barriers to the open source world, to open weight models or our local champions.

In short, this call for slowdown seems to me mainly a maneuver to ensure some stability in a market that has gotten out of hand. If the authors were really sincere about the concerns of model alignment, if they were worried about AI escape risks, they already have the means to work on the problem. The real existential risk today is more about their future IPOs, rising competition and a European market that could choose another path with open weight models. It seems unlikely in any case that China would subscribe to this call, nor would the Trump administration, and I hope Europe won't give in to the temptation to strengthen legislation at the cost of our future sovereignty.