14/09/2026
Is Dario Amodei and Antropic's plan to slow down (pace?) AI development going to work? The clue might be in the plan. All you have to do is leave your belief at the door and imagine the US and China collaborating. In the next handful of years.
Admittedly, I'm getting ahead of your here, because global collaboration is step three in the plan, so there is every chance the first steps are sufficient. What are they?
Step one is what Dario calls embedded evaluators.
Catchy.
Anthropic will be 'unilaterally committing' to third parties who have employee-like access to verify safety practices and report incidents. They will be able to check whether Anthropic 'is actually following the training, deployment, operational, and safeguards practices they claim to be following.'
Step two is 'Pacing with the democracies.' If I had read this three years ago I would have laughed out loud and sent you a gif. In 2027 it actually doesn't sound as unhinged. A lot of preparation has been done, after all.
"The most effective method of pacing is via regulation that targets all US frontier AI companies," says Dario in his blog. "That covers even those who are unwilling to cooperate voluntarily."
He then goes on, "I believe all frontier labs should partner with government to formalize the idea of permanent embedded evaluators to better prevent and document internal alignment incidents like those that have occurred in the last few months, and to implement regulation focused on keeping capabilities in balance with safety."
Then we get to step three. Global Pacing. Here Dario In parallel with pacing within democracies, we should also aim for a worldwide pacing of the frontier, though this will be much harder to achieve. Global pacing will require cooperation with China, the autocratic country with by far the most advanced AI capabilities.
Do you think it will work? Or is it too little, too late?
