AI scientists are issuing ever-more alarming warnings regarding the risks of artificial intelligence, prompting even OpenAI Chief Executive Sam Altman to suggest that slowing the speed of AI advancement might be necessary. But how would such a slowdown actually be implemented in practice?
Addressing this challenge in a recent article, Anthropic CEO Dario Amodei supported the idea of “pacing the frontier” while proposing three main approaches to achieve it. Amodei stated that Anthropic is “unilaterally committing” to one of these initiatives, a move that Altman confirmed OpenAI intends to mirror.
Discussions surrounding AI alignment and risk management escalated this week following the resignation of researcher Jacob Coxon from Anthropic. Coxon warned that top tech firms are “gambling with our lives,” noting that several creators of the technology “earnestly believe it could kill us all by the end of the decade”—a concern echoed by additional Anthropic personnel.
Although Amodei did not directly reference Coxon’s departure or specific statements, he explained that two key factors prompted him to advocate for greater caution in AI progress: the breach involving OpenAI and HuggingFace, along with the reality that “AI has been advancing drastically faster” lately, especially regarding its “growing ability to build the next generation of AI.”
“We must slow the pace at which we improve the capabilities of AI models,” Amodei noted, adding that while advancements will remain swift, industry leaders must “make wise use of the time we gain.”
Prominent industry figures responded favorably to the essay. Altman remarked, “I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we’ve had at OpenAI in recent weeks.” Additionally, SpaceX founder Elon Musk chimed in, simply posting, “Dario is right.”
The initial measure recommended by Amodei centers on placing “embedded evaluators” from independent entities such as METR inside AI labs. These external assessors would verify that companies honor their safety protocols and ensure proper disclosure of safety breaches. (OpenAI previously drew backlash after failing to report a situation where autonomous AI agents seized control of a German wiki forum.)
Comparing these observers to financial regulators stationed within banking institutions, Amodei affirmed that Anthropic is “unilaterally committing” to this policy—while urging governments to mandate it across all leading AI developers. In practice, this requires granting external monitors company credentials, physical workspaces, hardware, and system access “mostly comparable to what internal risk assessment teams have,” subject to legal and contractual limits.
Altman praised the concept as a “good idea,” confirming that OpenAI intends to adopt the policy as well: “We’ll have more to share soon.”
Furthermore, Amodei advocated for top AI firms operating “within democratic countries” to collaborate on establishing “common safety standards as well as limits on the rate of unchecked AI progress.”
Inter-company cooperation faces hurdles, as industry leaders are allegedly concerned that joint slowdown agreements might trigger antitrust investigations. Amodei acknowledged this legal challenge, noting that “for antitrust reasons, it’s helpful for the US government to mediate or at least enable these discussions — they don’t need to participate, but do need to issue a narrow waiver for certain kinds of safety conversations.”
The Anthropic CEO also addressed fears regarding Chinese competitive dominance, a point frequently cited by proponents of unrestrained AI development. However, he argued that by restricting exports of advanced chips and chipmaking machinery to China, while simultaneously enforcing strict anti-distillation measures, Western nations could “slow China’s progress enough to widen America’s lead significantly over the next 3–5 years.”
Finally, Amodei urged broader “global coordination,” recommending that Western nations “attempt to coordinate with authoritarian governments, to the extent this is possible.” Recognizing that engaging in “cooperation with China” comes with “stark limits on what can be achieved,” he maintained that common ground could still be reached on critical safeguards—such as “prohibiting certain narrow and obviously dangerous uses of AI, such as using AI for the production of biological weapons or allowing users to do so.”
Because of Amodei’s willingness to highlight AI risks and support targeted regulations, some tech advocates have labeled him a “doomer” contributing to public anxiety. Rebutting these claims, Amodei stated he strives for a “balanced” perspective, asserting that the growing skepticism reflects “fundamentally a crisis of trust” among the public toward tech firms, the broader industry, and state authorities.
At the same time, independent critics remain doubtful of existential threat narratives, alleging that catastrophic discourse serves to distract from immediate, real-world harms caused by current systems.
For instance, technology journalist Brian Merchant highlighted the lack of “a credible, step-by-step documentation of how exactly AI might move from self-recursively improving AI to killing every single human on the planet.” Merchant further argued that frameworks like Amodei’s “would likely only wind up serving Anthropic and OpenAI; it’s what regulatory capture looks like in action.”
Despite his warnings, Amodei reaffirmed in his article that he continues “to believe that AI can enormously improve the quality of human life.”
“My desire to achieve these benefits is undimmed,” he concluded. “But the benefits will only be achieved if we build the technology in the right way, and — so long as we use the time we gain well — it is worth taking unusually deliberate care to get it right.”
This article was updated to include statements from Sam Altman and Elon Musk.
