Anthropic CEO's Plan to Slow AI Development: 3 Strategies

The Growing Calls to Pace AI Development

For years, warnings about AI risk came mostly from a fringe of researchers. That has changed. Increasingly dire warnings now come from inside the labs building the technology, and the conversation has shifted from whether to slow down to how.

OpenAI CEO Sam Altman has said it may be time to "pace" AI development. The word is deliberately vague, and it raises an obvious question: what would pacing actually look like in practice?

Pacing is not the same as stopping. It could mean delaying the release of a model until safety evaluations are complete, capping the compute used in a training run, or coordinating release schedules across labs so no single company races ahead. Each approach carries trade-offs, and each is difficult to enforce when competitors are advancing quickly.

The debate has moved quickly from abstract warnings to concrete proposals, as the sections below show.

Amodei's Three Strategies for Pacing the Frontier

In a new blog post, Anthropic CEO Dario Amodei echoed the call to "pace the frontier" and outlined three broad strategies for doing so. His proposals aim to slow the most advanced AI development while still allowing progress to continue.

Amodei's three strategies are:

1. Transparency. Frontier labs should openly share information about their capabilities and safety practices.
2. Coordination. Developers should work together on shared standards rather than racing unilaterally.
3. Governance. Public institutions should play a role in setting the rules that govern frontier development.

Together, these strategies reflect a view that pacing the frontier is not about halting progress, but about steering it deliberately. Amodei's framing places responsibility on both labs and governments to act before risks outpace oversight.

For more on how these ideas translate into practice, see the Anthropic news page.

Anthropic's Unilateral Commitment

Amodei said Anthropic is "unilaterally committing" to one of the three strategies. That means the company is not waiting for competitors, regulators, or the wider industry to agree on a common pace before acting. Instead, it is choosing to hold itself to a specific commitment on its own.

The word "unilaterally" is significant here. In a race where rivals are pushing forward, a single lab restraining itself risks being overtaken. Amodei's framing acknowledges that tension but presents the decision as a deliberate choice rather than a coordinated one.

This approach places the burden of pacing on Anthropic itself, rather than on an external body or a collective agreement. It also sets a public benchmark: the company can be measured against the commitment it has made.

For observers of the AI safety debate, the move is notable because it tests whether voluntary, go-it-alone restraint can work when competitive incentives pull in the opposite direction. Whether other labs follow will shape how much this commitment matters in practice.

OpenAI Follows Suit as Safety Debate Intensifies

Sam Altman chimed in to say OpenAI will follow suit. The chief executive added his voice to a growing chorus of lab leaders signalling that frontier AI development may need to slow down, echoing the commitments made by Anthropic earlier in the week.

The debate over AI safety and alignment intensified this week after researcher Jacob Coxon wrote that he's worried about the trajectory of frontier model training, arguing that capability gains are outpacing the field's ability to verify that systems remain aligned with human intent. His post quickly circulated among researchers and policymakers.

The exchange highlights how quickly the conversation has shifted. What began as scattered calls for restraint has become a public back-and-forth among the very labs building the most capable systems. Whether these statements translate into binding commitments — or remain voluntary gestures — is the question now facing the industry.