Top AI Builders Say: Let's Slow Down a Bit
Sam Altman, Dario Amodei, Demis Hassabis and Elon Musk have loosely agreed over the weekend to slow down AI development to "pace the frontier" The Verge.
Altman runs OpenAI, Amodei runs Anthropic, Hassabis helped start Google DeepMind, and Musk runs SpaceX. The agreement is loose. It is not a contract, a joint venture, or a product delay.
The plan comes from Amodei. In September 2026 he published an essay called "We Must Pace the Frontier" Dario Amodei. It lists three steps. Altman, Amodei, Hassabis and Musk have at least partly backed outside auditors, rules for labs in the U.S., and a global slowdown deal.
Amodei asked AI companies to slow how fast models gain new abilities Reuters. That means how quickly AI gets smarter, not how many people use it. Anthropic separately said development should slow down and be closely watched. Amodei also said earlier calls to pause AI, floated since 2023, made "little sense" at that time CNBC. He described pacing as a practical operating choice, separate from past pause letters, not a full stop.
Support followed in public posts. In September 2026 Altman posted on X in support of the essay and the need to "pace the frontier". In September 2026 Musk posted support for the proposal The Guardian. Senator Bernie Sanders said on X that Amodei, Musk and Altman agree on slowing AI to "pace the frontier".
Inside OpenAI the talk started earlier. Altman told staff at a company meeting that OpenAI was open to slowing development. He said pacing the frontier had been a main topic at OpenAI in recent weeks.
The broader context here is that the three steps are very different jobs. Outside auditors would need access to models and test results and agreement on when to pause, like safety inspectors on a factory line. National rules would need licenses, reporting and enforcement. A global deal would need checks across countries with different interests. If new base models arrive less often, work would shift to testing, fixing and safely running current systems.
In my view, doubts about enforcement make sense, but the technical idea is healthy. Rival labs have rarely stuck to voluntary limits, and audit rules are still patchy. If a slower pace holds, safety work, better testing and normal business use would have time to catch up. That kind of consolidation has usually made new technology more useful in the end.
