Should the DOD (DOW) control AI?
Inside the decisions being made tomorrow about our future
Jack Clark is a co-founder of Anthropic. Two days before this interview posted, the Pentagon gave his company a deadline: remove your AI safety guardrails or be designated a supply chain risk — the same designation reserved for foreign adversaries like Huawei.
Clark sat down with Ezra Klein anyway. For 98 minutes. And he talked.
He talked about what happens to entry-level jobs when AI agents can complete entire workflows autonomously. He talked about recursive self-improvement and what it means that there are no visible signs of the technology plateauing. He talked about what it means that the same tools reshaping the economy are also being acquired — under pressure — for military use with no restrictions.
The chapter at 1:20:08 is titled ‘Anthropic’s dispute with the Department of War.’ Not Defense. War. Clark named it that himself.
Here’s what makes this interview different from the usual AI hype cycle: Clark doesn’t perform optimism. He’s one of the architects of this technology, and he talks about it the way someone talks about a force they respect and fear in roughly equal measure. He uses the phrase ‘global safety monitoring’ without irony. He talks about parenting in the age of AI not as a marketing question but as a genuine moral one.
Most AI coverage asks: how fast? This interview asks: fast enough that we can govern it? And the answer Clark gives is uncomfortable.
Watch it. The full 98 minutes if you can. Skip to 1:20:08 if you want to understand what’s happening right now, today, between the company that makes the AI you’re probably already using and the people who want to use it without limits.