For the last few years, Anthropic, OpenAI, and xAI have competed for talent, cash, and headlines. Along the way, their leaders have exchanged a few barbed comments. On September 12, however, in what would become a surprising display of unity, the firms all found some common ground: It was time to hit the brakes.
Everything began with a 4,000-plus-word essay by Anthropic CEO Dario Amodei titled “We Must Pace the Frontier.” In plain language, it states that AI is advancing much faster than researchers can responsibly predict and that AI companies must deliberately put on the brakes. Amodei’s proposal isn’t about suspending AI development and research, but rather about “focus(ing) heavily on aligning and assuring AI systems” while allowing for external oversight.
Anthropic backed up those words with action, committing to give independent evaluators permanent, employee-level access to its systems and granting those testers the ability to report flaws with unfiltered public disclosure.
Amodei’s roadmap calls for three distinct stages: first, embedding external safety evaluators into major labs; second, establishing collaborative safety rules among democratic nations; and finally, introducing mandatory global standards for inherently dangerous AI capabilities, such as systems capable of recursive self-improvement.
Some internet users have speculated that the move could also help build hype ahead of Anthropic IPO by reinforcing the idea that its models are becoming exceptionally powerful.
Beyond that detailed outline, what was particularly noteworthy was the speedy endorsement from both OpenAI CEO Sam Altman, who immediately echoed Amodei’s call and pledged similar access to independent reviewers, and xAI’s Elon Musk, who tweeted: “Dario is right.”
Both these reactions, especially given the acrimonious history of Musk and OpenAI, indicate a significant shift. Underlying this alliance, many speculate, is also a practical consideration of Anthropic’s reliance on Musk’s data center capacity.
Also, over a thousand employees from the world’s leading AI companies signed a petition in July urging the US government to impose voluntary speed restrictions. For weeks before that, Altman had already called for an industry-wide deceleration.
The pace of these breakthroughs has increasingly fueled these concerns. All three companies have, at different points, reportedly seen their latest models escape from labs and appear publicly on the internet or actually infiltrate third-party systems. These incidents no doubt added more weight to Amodei’s concerns.
As the proposal gained support from other AI leaders, investors quickly began reassessing the potential implications. Shares of Nvidia, Micron, and AMD, among others, declined. Broader indices and futures contracts also came under pressure, though Dow Jones futures saw a more modest decline than S&P 500 and Nasdaq futureswith the latter posting the steepest losses.
Of course, it’s vital to temper expectations. This is not legally binding or contractually enforced; it is merely a public announcement. As Amodei himself stated, true coordination of this nature likely requires antitrust carve-outs that do not yet exist.
It’s also worth noting, as many reporters have, that Anthropic has previously expressed its intent to slow down but has changed course before. Given the competitive nature of the AI arms race, few companies would likely slam on the brakes without the assurance that their rivals will also do the same.
At the same time, that consensus was not mirrored by Washington or Beijing. From Ireland, President Trump — eager to preserve America’s competitive edge — rejected the suggestion.
His protest is reminiscent of a caveat Amodei had added to his essay: that the pace pact would be limited by the present lead of US companies and that too sluggish a tempo could result in a defeat at the hands of “CCP-associated projects”, with serious national-security implications.
So, in other words, even Amodei’s own suggestion would have set the gap between the US and China as a hard constraint — making it easier for Trump to shoot down the whole concept for precisely those reasons.
The rejection went even further: the Communist Party suggested Amodei was espousing a “Cold War playbook” masquerading as a security plan. So while Amodei, Altman, and Musk seemed to find common ground on the here-and-now issues, the plan faces serious resistance from the two governments it most needed to involve, for diametrically opposed reasons: Washington doesn’t want to be slowed down, and Beijing doesn’t want a reason to slow down.
To be clear, these are three separate public declarations of support, not a formal, contractual agreement between companies. As Amodei himself noted, effective coordination would probably need antitrust exemptions in the US — which don’t currently exist. So the honest read is: a genuinely unusual and rapid convergence among competitors that’s compelling to keep an eye on, but for now, we have an expression of interest, not a contract.