Anthropic co-founders Dario and Daniela Amodei have shifted from racing to lead the AI industry to advocating for deliberate slowdowns in model development, a dramatic pivot that's reshaping how the entire sector approaches safety and competition. Just four years after OpenAI dominated the AI landscape with ChatGPT, the brother-and-sister duo have guided Anthropic to surpass their rival in both valuation and revenue, powered by their Claude models and a specialized coding version called Claude Code. Yet rather than accelerate further, they're now calling on the entire industry to pump the brakes.

The Amodeis' influence extends far beyond their company's technical achievements. Dario Amodei published an essay on September 12 proposing a three-step plan to intentionally pace AI development without sacrificing the United States' competitive edge. The proposal garnered immediate support from two fierce industry rivals: OpenAI CEO Sam Altman and Elon Musk, marking an unusual moment of consensus among leaders who typically compete fiercely.

What Changed to Make AI Leaders Embrace Caution?

For years, calls to slow AI development were largely dismissed as impractical. Dario Amodei acknowledged this shift in his essay, explaining that the case for pacing "made little sense" back in 2023, when models lacked the capability to take real-world actions or engage in sophisticated deception and cyberattacks. Today's models are fundamentally different. Claude Code can design, debug, and execute working programs faster and better than many human engineers, while Anthropic's more powerful Mythos model can identify thousands of critical vulnerabilities in widely-used software.

The geopolitical stakes have also intensified. In spring 2026, Anthropic clashed with the Pentagon over restrictions on military use of its models, and the Trump Administration subsequently placed export controls on Mythos and its sister model Fable 5. These tensions underscored that AI development is no longer purely a commercial matter; it's now entangled with national security and international competition.

How to Understand Anthropic's Three-Step Safety Plan?

• Third-Party Verification: Anthropic has unilaterally committed to granting independent evaluators employee-level access to verify safety practices and report incidents, a step Amodei says the company has already implemented.
• Coordinated Standards Across Democracies: The second step encourages leading AI companies within democratic nations to establish common safety benchmarks and practices, reducing the risk that competitive pressure forces corners to be cut.
• International Coordination: The third step calls for coordination between democratic governments and authoritarian regimes to prevent a dangerous race to the bottom where the most reckless actors set the pace.

Amodei stressed that 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". This framing is crucial: the Amodeis are not arguing for a freeze, but rather for deliberate, measured advancement with safety checkpoints built in.

Amodei

Why Is This Moment Significant for the Industry?

The alignment between Altman, Amodei, and Musk signals a fundamental shift in how the industry views its own trajectory. OpenAI's decision to postpone its long-anticipated initial public offering (IPO) until at least 2027, citing safety concerns, demonstrates that these aren't just rhetorical commitments. CFO Sarah Friar had told employees just weeks earlier that an IPO could happen in 2027 or sooner if the business continued to grow, but Altman's reversal shows that mounting safety concerns are now reshaping business timelines.

Pressure is mounting from multiple directions. Researchers within Anthropic itself have raised alarms; fellow Jacob Coxon recently resigned, saying the company and OpenAI are "gambling with our lives" and that AI builders "earnestly believe that it could kill us all by the end of the decade". Meanwhile, lawmakers in both parties are demanding new AI safeguards and calling for tech leaders to testify after cyberattacks carried out without direct human control. State and local officials are also confronting growing opposition to AI data centers and their demands on power and water resources.

"To be clear, 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," Dario Amodei wrote in his essay.

Dario Amodei, CEO at Anthropic

The Amodeis' willingness to advocate for industry-wide caution is particularly striking given Anthropic's commercial position. The company is preparing for what's expected to be a historic IPO, yet they're championing measures that could slow their own growth. Dario Amodei acknowledged the tension in a CNN interview, noting that moving too slowly could give authoritarian governments an advantage: "If we go too slow, I still believe that the wrong people will be in charge of the technology. And that, again, will bring the probability of things going wrong very high".

Dario Amodei

Anthropic's head of public policy, Sarah Heck, called on lawmakers to match the industry's commitment with regulatory action, including blocking sales of advanced chips to adversarial nations and enacting national laws requiring testing of frontier models with the power to block unsafe systems. This represents a notable shift: the Amodeis are not just asking companies to self-regulate, but actively inviting government oversight.

What Does This Mean for AI's Future?

The Amodeis' pivot reflects a broader recognition within AI research that the technology has reached a critical inflection point. In 2023, Dario Amodei was among prominent researchers and executives who signed a statement declaring that "mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war". What's changed is that models are now powerful enough to act on these concerns in concrete ways.

Anthropic's own research underscores this urgency. A recent paper by Anthropic fellow Chen Yueh-Han demonstrated that automated AI systems can reliably improve a model's alignment and safety performance, with the best automated methods outperforming experienced human researchers within six hours. While this capability could accelerate safety improvements, it also illustrates how quickly AI development is advancing and how difficult it may become for humans to maintain meaningful oversight.

The Amodeis' influence on this moment cannot be overstated. They've transformed Anthropic from an ambitious startup into a company that rivals OpenAI in market valuation and revenue, proving that a safety-first approach doesn't preclude commercial success. Now they're using that credibility to reshape industry norms, arguing that the path forward requires not just better technology, but better governance and deliberate restraint. Whether their three-step plan gains traction will likely determine whether AI development proceeds as a coordinated, safety-conscious effort or devolves into an uncontrolled race where the most cautious actors are left behind.

" }