Why One AI Leader Says We Need to Slow Down AI
Anthropic CEO Dario Amodei called on September 12, 2026, for the AI industry to slow down. He announced an essay titled We Must Pace the Frontier in a social media post, according to The Guardian.
The announcement was a post on the @DarioAmodei account on X at 2:01 PM. It linked to darioamodei.com/post/we-must-pace-the-frontier. That post had 4.8 million views, per the X post.
Amodei offered a three-part plan to slow AI progress. Anthropic said it will follow the first step on its own. That step is about outside checks on safety. It would give outside experts permanent access, like employees, to Anthropic systems. They could check safety steps, report problems, and test whether models act as intended while they are still being trained.
Amodei wrote that AI abilities are being improved too fast. Safety work needs time to catch up. He has said AI will pass human thinking ability on most tasks within a "small number of years." Former Anthropic researcher Jacob Coxon warned that AI could cause human extinction by 2030.
There is deep distrust between Amodei and OpenAI CEO Sam Altman, as reported by Reuters. An official identified as Lutnick sent a letter to Amodei ordering the company to suspend export of its AI models, according to Reuters.
The broader context here is a hard group problem layered over a checking problem. One company can set an example. That does not remove pressure on others to move fast. Rival labs face demands from investors, pressure linked to national security, and fear of falling behind. Amodei's answer seems to be openness that can be checked. Permanent access would close the knowledge gap between builder, checker, and later the government. It is like letting factory safety inspectors work inside every day instead of visiting once. It also tests whether checks during training find problems earlier than tests only before release.
Looking at what this means for rules, two questions matter. The first is whether other leading labs will give the same access. If they do, it could help build shared checks or inspection ideas tied to future deals between countries. If they do not, Anthropic pays the cost and hassle alone. Trust is scarce. The reported distrust between Altman and Amodei shows how personal and company splits make any joint slowdown hard, even when the stated risks overlap.
A second question for the future concerns the state. Orders to suspend exports are blunt tools. They control where models go. Checker systems control what is known about how models are built. The two can work together or clash. Governments that want control may like deeper records. Companies that want freedom to release products may find openness brings more government action. For people who follow global institutions, the question is which approach can work across borders, where laws, technical skill, and national interests differ sharply.
