Amodei sparks backlash with call to slow AI scaling

Dario Amodei published a 3,000-word manifesto asking competitors to slow down. The backlash was instant.

The Anthropic CEO proposed three immediate steps: embed third-party safety evaluators directly inside frontier labs, establish common standards across democratic nations, and coordinate limits internationally. OpenAI chief Sam Altman, Google DeepMind lead Demis Hassabis, and Elon Musk backed the initiative within hours. Amodei cited recursive self-improvement and recent multi-agent cyberattacks as the primary drivers requiring external oversight.

Critics outside the closed labs saw a commercial survival tactic rather than existential panic. Anthropic and OpenAI face crushing capital expenditures and mounting losses ahead of expected initial public offerings. Investor Michael Burry observed that halting expensive frontier training allows labs to cut spending without surrendering relative market share. On Breaking Points, Krystal Ball noted that existing commercial products boast 80 percent profit margins once training stops. That shift gives labs a clear way to show sudden profitability to Wall Street.

The framing transforms an operational bottleneck into a moral crusade.

"Self-policing will not protect the public interest."

- Krystal Ball, Breaking Points with Krystal and Saagar

Politicians and policy advisers across the spectrum rejected the self-regulation scheme. On All-In, Vice President J.D. Vance rejected Amodei’s call for central regulatory control. Vance argued that if labs build dangerous systems, the burden sits on them to fix the software rather than petitioning Washington for global authority. Former President Donald Trump called AI doom narratives an economic hoax designed to stifle data center construction. White House AI adviser David Sacks argued that established labs already operate as a duopoly and do not deserve antitrust exemptions to coordinate research speeds.

Antitrust experts and venture capitalists warned that joint calls for training pauses resemble illegal collusion under competition law. On All-In, David Sacks and David Friedberg argued that the viral panic originates from effective altruist advocacy networks trying to manufacture a federal AI regulator. Such a regulator would ban open-source software by making unmonitored model weight distribution illegal. That outcome locks in a permanent moat for closed incumbents while stripping developers of free tools.

"The doomer narrative relies on fear rather than evidence."

- David Sacks, All-In with Chamath, Jason, Sacks & Friedberg

Nvidia CEO Jensen Huang dismissed existential risk predictions as unscientific fearmongering during an interview on All-In. Huang pointed out that past doomsday predictions failed completely, from the total elimination of radiologists to automated coding claims. He noted that 80 percent of AI startups build on open-source foundation models. Winning the global AI race requires deploying software across the real economy as fast as possible, not pausing research.

Geopolitical friction makes any coordinated U.S. slowdown effectively impossible. Chinese state media condemned Amodei’s essay as a hypocritical attempt to protect American monopoly power while China advances its open-weight models like DeepSeek. DeepMind economist Alex Zhus noted that while moderate pacing gives enterprise buyers time to integrate tools without sudden obsolescence, national security priorities in Washington and Beijing will ultimately override corporate safety agreements.

The AI race will not pause for voluntary charters.