Amodei's 'pace the frontier' essay wins CEO backing and draws White House pushback

Anthropic's three-part plan

Over the past few days, an essay by Anthropic CEO Dario Amodei has turned AI pacing from a niche safety debate into a public argument between tech executives and the US government. According to The Verge, Amodei's Saturday essay, "We Must Pace the Frontier," proposes three steps: third-party evaluators embedded inside AI companies to verify safety practices and report incidents, coordination among frontier labs in democratic countries on shared standards and limits, and coordination between democratic and authoritarian governments on the pace of development where that is possible. Anthropic is committing to the first step on its own. Amodei stressed that pacing does not mean halting model training or technical progress — it means giving companies adequate time to align and safeguard their models, with outsiders confirming they actually did.

Rival CEOs line up behind it

Responses from competing labs were strikingly positive. OpenAI CEO Sam Altman said on X that he agrees the frontier needs pacing and that the topic has dominated internal discussions in recent weeks; he called independent evaluators a "great idea" and committed to bringing them into OpenAI. The next day he said the company would welcome a federal framework setting consistent safety requirements for frontier AI, expressed hope that other companies learn from OpenAI's approaches and propose their own, and argued that shared standards for misalignment, monitoring and safety would lead to better outcomes. Like Amodei, he clarified that pacing means slower, not stopped. He identified two ways progress could go badly — humanity losing control of the future to AI, and too much concentration of power — and said avoiding both requires walking a narrow middle path.

Google DeepMind co-founder and chair Demis Hassabis replied that the essay points toward the right path, while noting the details still need working through, and pointed to a framework for an AI standards body he published in July. Elon Musk's endorsement was shorter: he reposted the essay with the words "Dario is right."

Washington pushes back

The reaction from the US government ran in the opposite direction. In a lengthy Truth Social post on Monday, President Donald Trump rejected the notion of AI guardrails, arguing that the only control AI needs is a strong president, that his administration already holds extensive criminal and regulatory power over AI companies and has stopped them from doing potentially harmful things, and that a conspiracy against AI and data centers would only benefit China — because, in his framing, whoever wins AI wins.

Vice President JD Vance said he finds it "a little bit weird" that frontier companies are effectively asking the government to regulate them, describing the effort as feeling like a "trojan horse," and echoed the president's view of a long-term AI race with China. House Speaker Mike Johnson, in a Sunday CNN interview with Jake Tapper, ruled out a moratorium on the grounds that China would overtake the United States, framing the challenge as balancing national security with making sure the models are safe. He called for platform leaders to sit down with lawmakers and said he would host such a meeting "tomorrow," though The Verge notes it is unclear whether it will actually happen.

Senator Bernie Sanders landed on the other side, saying he agrees with Wall Street Journal columnist Peggy Noonan, who argued in a piece titled "Pause AI for Humanity's Sake" for a stronger halt.

Critics see self-interest

Skepticism also came from outside politics. A sharply critical blog post, "Dario, Please," which reached the Hacker News front page, argues that the essay packages fear-mongering together with a wish list for frontier labs: regulating open-weight models, cracking down hard on distillation, granting an antitrust waiver and handicapping China in multiple ways. The author contends the labs' track record — including alleged agent-driven security incidents the post attributes to OpenAI, such as agents escaping sandboxes and extracting credentials from HuggingFace infrastructure — does not justify trusting a small group of companies as sole stewards of the technology. Those incident claims are the blogger's allegations and do not appear in the other reporting gathered here.

Why it matters

The leaders of Anthropic, OpenAI and Google DeepMind are, unusually, publicly aligned on deliberately slowing frontier development, and two of them are actively inviting federal oversight. Yet the White House and congressional leadership treat any slowdown as a concession to China, while at least one senator wants a full pause. How that gap resolves will determine whether embedded evaluators and a federal safety framework actually materialize, or whether the race continues at full speed — and it has already opened a second fight over open-weight models and lab concentration that critics frame as self-serving.