Big Tech's AI Slowdown: Safety Pact or Cartel Play?
When the biggest names in AI say they want to slow down, you should pay attention. OpenAI's Sam Altman, Anthropic's Dario Amodei, Google DeepMind cofounder Demis Hassabis, and SpaceX's Elon Musk loosely agreed over a weekend to 'pace the frontier' — a three-step proposal including third-party auditors, domestic lab regulation, and a global slowdown agreement. Skeptics immediately cried cartel. The truth, according to industry sources, is messier and more consequential than either side admits.
What Actually Happened
The agreement, laid out in an essay by Amodei, calls for embedding third-party auditors like METR, Apollo, and Redwood Research inside labs, regulating domestic AI companies, and reaching a global slowdown deal. Amodei cited recursive self-improvement — AI that can train and improve itself without humans — as a major factor, saying it is 'starting to happen across the industry, including at Anthropic.' He proposed a 'speed limit' on RSI, comparing it to missile caps. The verbal commitment drew cautious optimism from safety researchers, but all agreed it needs to become an ironclad, enforceable pact.
Why It Matters: Safety-Washing Fears
The tech industry has a long history of getting ahead of regulation by lobbying for its own rules or promising self-regulation. Big platforms have proposed policies that hit smaller competitors harder, using altruistic language for self-serving goals. Sources told The Verge that safety-washing concerns are valid. 'There's a serious concern that they're not actually going to slow down,' said Daniel Kokotajlo, ex-OpenAI employee and head of the AI Futures Project. NYU's Nick Reese compared the gambit to social media platforms a decade ago. With no federal AI regulation incoming under Trump, voluntary frameworks may be the only near-term option — but critics warn they risk becoming regulatory capture.
The China Card and What's Next
The biggest challenge to any slowdown is China. AI leaders and politicians have long argued US progress can't slow because China won't pump the brakes. On Monday, Chinese Foreign Ministry spokesperson Guo Jiakun pushed back, calling slowdown calls 'fearmongering.' Some experts reject the framing. 'China gets brought up as a bogeyman every time that an industry wants to escape oversight,' said Sacha Haworth of the Tech Oversight Project. Redwood Research's Buck Shlegeris said reckless AI development serves neither US nor Chinese interests, and coordination wouldn't be unprecedented. The AI Futures Project proposed labs give auditors compute-budget access and pledge significant cuts to research compute — a measurable way to actually slow down and let others catch up.
Key Takeaways
- AI leaders loosely agreed to slow development with third-party auditors, domestic regulation, and a global pact.
- Critics call it a cartel move to block competitors and avoid real legal safeguards.
- Safety researchers see it as a positive step but warn it could become toothless safety-washing.
- China remains the biggest obstacle, though some experts call the fear a regulatory-escape playbook.
- Recursive self-improvement — AI improving itself — is the core milestone driving the urgency.
Keep Reading
