One AI Leader Says We Should Slow Down AI and Check It More Carefully
Anthropic CEO Dario Amodei says the most powerful AI systems should be built more slowly and checked by independent outsiders.
His three-step plan calls for outside safety experts to work inside AI labs, for U.S. companies to follow shared safety practices, and for countries to reach international agreements, possibly with help from governments The Verge. Anthropic has described the issue as pace. The company said AI development should slow down and be closely watched BBC.
What Amodei proposed
The first step is stronger safety testing of advanced models Reuters. The second step is about access. Amodei called for independent experts to work inside labs with employee-like access to check safety practices Reuters.
That would change how testing works. Right now, outsiders mostly test a finished model by asking it questions. They cannot see how it was built or what safety steps were used. Think of it like food inspectors working inside a kitchen, not just tasting the meal at the table. The third step moves from single labs to joint action, first in the U.S. and then between countries.
Amodei paired the plan with a warning. He said that within six to 12 months AI could be able to lead a swarm of agents able to take over the entire internet AP. He did not present that as a research result. He presented it as a reason to slow down now.
Amodei also wrote that Anthropic "has always advocated for well-considered regulation of AI" The Verge. The plan is only a proposal. No lab has adopted it, and no government has put it into effect.
How rival labs responded
Other AI leaders reacted in different ways. OpenAI CEO Sam Altman, Google DeepMind co-founder Demis Hassabis, and SpaceX CEO Elon Musk publicly agreed with parts of the three ideas The Verge. Their agreement was partial. It was not full support for a slowdown.
Meta CEO Mark Zuckerberg opposed limits on a company's freedom to direct AI work The Verge. Zuckerberg wrote that "every lab has the responsibility and incentive to move at the pace required to train its models safely" The Verge.
The split is narrow but important. Both sides say they care about safety. One side says outside checks and teamwork are needed. The other side says each lab can judge its own pace.
OpenAI added a separate point. Policy chief Chris Lehane told The Verge that government has an important role in AI safety, along with steps taken by companies. Lehane also said there is growing bipartisan support for mandatory national safety standards for frontier AI The Verge.
Washington and an industry regulator that stalled
The White House rejects the premise. The Trump administration called the idea of a safety crisis in AI a "hoax" The Verge.
An earlier industry idea also failed. The Wall Street Journal reported that Zuckerberg, Musk and Nvidia CEO Jensen Huang blocked a plan for an industry-funded, independent AI regulator modeled on FINRA The Verge. FINRA, the Financial Industry Regulatory Authority, is a private group that oversees stock brokers under government supervision.
The broader context here will be familiar from past technology shifts. Companies often share basic standards willingly. They rarely agree to slow down or pay more if rivals might not.
In my view, the key part is the plan for inside experts. Tests from the outside can check answers. They cannot prove what safety steps were followed, what risks were found, or whether the shipped version matches the tested version. Inside access would help. It would also raise hard questions about secrets, security and independence that are not yet answered.
Looking at what this means for builders, two tracks are running at once. One is practical, about better tests and clearer proof of safety. The other is political, about whether national rules move forward. Even without new laws, outside scrutiny is likely to grow. That will help labs that can show, in a way others can verify, that they built carefully.
