AI Slowdown Pact: Safety Move or Cartel to Freeze Out Competitors?

Over one weekend, the heads of OpenAI, Anthropic, Google DeepMind, and SpaceX reached a loose agreement to slow down AI development. Their stated rationale is safety: third-party auditors, domestic lab regulation, and a global pace-setting deal. But critics were fast to point out that the same agreement would also make life very difficult for smaller labs and open-source projects, and some are calling the arrangement an outright cartel. The story is still developing, and the full details have not been disclosed.

What happened

Detail	Fact
Who agreed	Sam Altman (OpenAI), Dario Amodei (Anthropic), Demis Hassabis (Google DeepMind), Elon Musk (SpaceX)
When	Over one weekend (exact date not disclosed)
Stated goal	“Pace the frontier” of AI development
Proposed mechanisms	Third-party auditors, domestic lab regulation, global slowdown agreement
Critic label	“Cartel”

Four of the most powerful figures in AI signed on, at least loosely, to a proposal that would embed independent auditors into leading labs, create regulatory controls for domestic AI development, and push for a coordinated global slowdown. The agreement was informal, reached over a weekend, and its precise terms have not been made public.

The four executives represent companies that together account for a large share of frontier AI research. OpenAI makes GPT-4o and o3. Anthropic makes Claude. Google DeepMind sits behind Gemini. Musk leads SpaceX and xAI. That these four names reached any kind of consensus is notable in itself, given their history of public disagreements.

Why do critics call it a cartel?

The concern is straightforward: a slowdown agreement between the biggest players in a market is, structurally, a way to protect incumbents. If development slows across the board, the companies that already have large models, large datasets, and large compute budgets stay ahead. Smaller labs and open-source projects, which rely on rapid iteration to close the gap, get frozen out.

Critics also argue the proposal avoids real legal accountability. Voluntary audits and soft international agreements are not the same as binding safety law. Some observers say the framing of “safety” is being used to achieve competitive goals without the scrutiny that antitrust law would normally bring. Our earlier coverage of why Altman, Amodei, and Musk aligned on slowing AI walks through the political backdrop to this shift.

Is the safety argument credible?

It is not obviously wrong. Frontier AI models are becoming more capable quickly, and the case for some form of external oversight is serious. Third-party audits, if designed properly, could surface risks that internal safety teams miss or have incentives to downplay. A global agreement, if it had teeth, could prevent a race to the bottom between countries.

The problem is that “pacing the frontier” sounds different depending on who defines the pace. If the four companies setting the pace are the same ones benefiting from it, the safety rationale and the competitive rationale point in exactly the same direction. That does not mean safety is insincere, but it does mean the incentives are badly tangled.

Our take

Both things can be true at once. These executives probably do have genuine concern about AI risks. They probably also see regulatory capture as a better outcome than open competition. History is full of industries that wrote their own safety rules and called it responsibility.

For business operators, the practical question is simpler: if a slowdown takes hold, the current generation of models may be the baseline for longer than expected. That makes it worth building workflows and processes around what exists today rather than waiting for the next leap. Our AI integration work is already focused on this: making current models genuinely useful, not betting on hypothetical future ones.

Watch what happens with open-source. If the agreement puts pressure on open-weight model releases, that is the clearest sign that competition, not safety, is the primary driver.

What to do about it

1. Audit your current AI tool stack and identify which capabilities you actually use today.
2. Build automations around stable, available models rather than waiting for the next release.
3. Follow the open-source model ecosystem closely. Restrictions there will signal intent faster than any press statement.
4. If your business relies on AI for competitive advantage, monitor any proposed regulatory changes to domestic lab rules, since those rules could affect access and pricing for API users downstream.

The cartel question will be settled by what the regulation actually says, not by what these executives say about safety.

Source: The Verge · AI