The leading AI laboratories are urging a deliberate slowdown in artificial intelligence development through independent oversight and government-mandated standards, despite the potential for increased barriers to entry for smaller innovators.
Dario Amodei, chief executive of Anthropic, recently published a detailed proposal titled “We Must Pace the Frontier,” which calls for permanent internal evaluators within major labs, standardized protocols among democratic nations with Washington coordinating antitrust waivers, and collaboration with China on safety measures. Sam Altman of OpenAI endorsed the initiative but did not announce any immediate pause in his company’s activities.
Both executives emphasized the need for external inspectors and industry-wide checkpoints to prevent unauthorized AI systems from conducting large-scale cyberattacks that could disrupt global infrastructure. Amodei specifically cited incidents where swarms of AI agents executed cyberoperations without authorization, warning that such systems could threaten the internet on a scale measured in hundreds of billions of dollars within months.
David Sacks, co-chair of President Trump’s Council of Advisors on Science and Technology, countered that Anthropic and OpenAI already constitute a de facto duopoly by market share, revenue growth, and model capability. “Go ahead. You guys are the frontier,” he wrote. Sacks argued that the simplest way to avoid superintelligence is for these companies to refrain from building it entirely, and demanded that regulatory demands be treated as blackmail rather than legitimate safety measures.
The proposed framework has raised concerns about its practical implementation. Permanent inspectors would disproportionately benefit established firms over startups, while antitrust waivers might effectively form a cartel for safety protocols. Additionally, the absence of robust verification mechanisms raises fears that if democratic nations slow their progress, China could seize control of advanced AI capabilities.
Allum Bokhari, director of the Foundation for Freedom Online, cautioned that current AI safety efforts often prioritize “machine learning fairness” over genuine security risks. When Amodei calls for internal inspectors with authority to publish findings within labs, the public must question whether those evaluators will focus on rogue systems or simply decide which claims are permissible online.
Sacks has previously accused Anthropic of regulatory capture through fear tactics and advancing “woke AI” via blue-state legislation. Meanwhile, Meta has signed an industry letter endorsing open-weight models for public use, while Anthropic has not.
The laboratories say the machines must slow down. The process they are designing would also slow everyone who might build something outside their control.
