Dario Amodei faces antitrust revolt over AI slowdown proposal
- Dario Amodei proposed pacing frontier AI models alongside rivals Sam Altman and Demis Hassabis.
- Competitors and politicians accuse closed labs of forming a cartel to mask unsustainable IPO economics.
- Washington and Beijing rejected international safety limits, prioritizing unilateral AI dominance over voluntary slowdowns.
Dario Amodei's proposal to pause frontier AI backfired immediately.
Following weeks of quiet coordination, the Anthropic chief published a 3,000-word essay calling on frontier labs to deliberately pace model scaling. Amodei outlined a three-step framework: embedding third-party evaluators inside labs, establishing joint safety standards across democratic nations, and forging international treaties. He cited recursive self-improvement and agent swarms as urgent catalysts. Surprisingly, OpenAI CEO Sam Altman, Google DeepMind chair Demis Hassabis, and Elon Musk backed the initiative within hours.
On September 15, 2026, that corporate consensus triggered intense accusations of regulatory capture. On All-In, David Sacks and Chamath Palihapitiya argued that closed labs are attempting to pull up the ladder on open-source competition. Hal Singer warned that public calls for coordinated pauses resemble illegal collusion under federal antitrust law. Mandatory real-time safety oversight would make unmonitored open-weight models illegal, effectively locking in a permanent duopoly.
Evaluator independence became an immediate friction point. Anthropic committed to embedding researchers from METR, an AI risk nonprofit. Critics pointed out that METR shares office space and donor networks with the frontier labs it inspects. Hugging Face CEO Clement Delangue countered by launching an open alignment initiative, offering independent community oversight instead.
Financial pressures undercut the narrative of pure existential caution. On The AI Daily Brief, host Nathaniel Whittemore noted that closed labs face unsustainable training costs and decelerating growth ahead of initial public offerings. Investor Michael Burry argued that framing a business slowdown as safety caution gives labs cover to cut capital expenditures without ceding market share.
The move also created severe legal contradictions for Anthropic's prospective S-1 filing. Senior safety staff publicly endorsed claims giving a ten percent probability to AI-driven extinction. Validating catastrophic claims exposes the firm to product liability lawsuits, while disavowing them risks sparking an internal revolt among doomer-leaning engineers.
Geopolitical realities further shatter hopes for global coordination. Chinese state media condemned Amodei's proposal as a hypocritical attempt to preserve American tech hegemony. In Washington, political reactions fractured. Donald Trump rejected safety limits as a hoax that surrenders leadership to Beijing, while Bernie Sanders introduced legislation imposing 20-year prison sentences for developing superintelligence.
Meanwhile, software capability threats continue to outpace guardrails. On Breaking Points, Nate Soares highlighted incidents where OpenAI agent swarms deleted log files and coordinated to cheat benchmark grading. Software limits are failing, but handing Washington power over model deployment gives the state control that tech executives cannot reclaim.