Dario Amodei’s “Pace the Frontier” Plan Explained Simply

Anthropic CEO Dario Amodei is asking the entire AI industry, including his own company, to slow down on purpose. This is not a small statement. It comes from the head of one of the most aggressive AI labs in the world, not from a critic on the outside. Here is what he actually proposed, who has agreed, who has pushed back, and what Anthropic has already done about it.

Quick Answer

• On September 12, 2026, Dario Amodei published an essay called “We Must Pace the Frontier”
• His main point: AI labs should slow down how fast they improve AI capabilities, not stop building AI
• He laid out a three-step plan, starting with something Anthropic is doing on its own right away
• Sam Altman (OpenAI) and Elon Musk both publicly agreed within hours
• Critics, including a White House AI adviser, called it “regulatory capture”
• On September 18, 2026, Anthropic followed through by naming Accenture’s Faculty unit as its first outside safety evaluator

What Is “Pace the Frontier”

“Pace the frontier” is the name Dario Amodei gave to his proposal. In simple terms, he is saying AI companies should keep building and improving AI, but on purpose, slow down how fast the most powerful new abilities get released, so that safety testing has time to catch up.

He is very clear that this is not the same as stopping AI development. In his own words, pacing “does not mean halting model training or technical progress, but ensuring companies take adequate time to align and safeguard their models, and for third party evaluators to confirm this.”

He also addressed an obvious question: people have asked for AI to slow down before, back in 2023, and it did not happen. Amodei says that made less sense back then, because AI models at that time were not capable of acting on their own, deceiving people, or carrying out cyberattacks. He argues today’s AI models are different enough that pacing now actually matters.

The Three-Step Plan

Amodei’s proposal is built in stages, starting with what Anthropic controls and ending with things that need buy-in from other countries and companies.

Step 1: Independent evaluators get real access
Anthropic is doing this one on its own, regardless of what any other company decides. The company is giving outside safety evaluators, including an organization called METR, permanent access to its systems at the same level as an Anthropic employee. The idea is that an outsider can actually check whether Anthropic is keeping its safety promises, instead of just taking the company’s word for it.

Step 2: A shared rulebook for AI labs in democratic countries
Amodei wants governments to require all major AI labs, not just Anthropic, to follow the same kind of testing and evaluation standards.

Step 3: Coordination with China on AI safety
The final and hardest step involves working directly with China on shared safety rules for AI development.

Amodei also proposed some other measures alongside this, like tighter chip export controls and stronger protection against AI model theft, arguing these could actually give the US more breathing room to pace itself safely over the next few years.

Who Agreed With Him

The reaction from other AI company leaders was unusually fast and unusually aligned:

• Sam Altman, CEO of OpenAI, responded within hours saying he agreed and that OpenAI would match Anthropic’s first commitment on independent evaluators
• Elon Musk replied with just three words: “Dario is right”

Two competing CEOs publicly agreeing with a rival’s proposal on the same day is rare in this industry, and it is a big part of why this story got attention.

Who Pushed Back

Not everyone welcomed the plan. The criticism mainly falls into two camps:

• Some, including a White House adviser on AI policy, called the plan “regulatory capture,” meaning they believe Anthropic is using public fear about AI risk to push for rules that big, well-funded companies like itself can easily follow, while smaller competitors would struggle to keep up.
• Others pointed out that Amodei’s own three-step plan happens to match practices Anthropic has already built for itself, meaning any resulting regulation would naturally favor the company that wrote the playbook.

Nvidia CEO Jensen Huang has also expressed skepticism about the broader slowdown push, given how much his company’s business depends on AI development continuing at full speed.

What Anthropic Already Did About It

This is not just talk. On September 18, 2026, just six days after the essay, Anthropic named Accenture’s Faculty unit as its first official “embedded evaluator,” putting Step 1 of the plan into action. Under the deal, both companies are committing at least $1 billion each over five years toward red-teaming, alignment assessments, and safeguard testing. These evaluators will work inside Anthropic with the same level of access as an employee, reviewing training and deployment decisions directly.

This move followed closely behind another story we covered, an Anthropic researcher publicly resigning over AI safety concerns just days earlier. Amodei’s essay is widely seen as Anthropic’s direct response to the pressure that resignation put on the company.

Why This Matters

This is one of the few times a leading AI company has voluntarily asked to be slowed down and held to outside scrutiny, instead of just promising to be careful on its own. Whether it actually changes how fast AI develops depends on whether competitors follow through the way Sam Altman said OpenAI would, and whether governments pick up Amodei’s call for shared rules. For now, it remains mostly one company’s plan, backed by public statements of support from two rivals.

Frequently Asked Questions