The U.S. Department of War's war on Anthropic
What ought to be the biggest news story right now
While I do frequently comment on current events in AI on this Substack, I am not doing it sufficiently regularly that readers can rely on it for keeping up with major events in AI. For that, my main recommendations would be Zvi Mowshowitz’ Don’t Worry About the Vase, and Transformer.
But I simply cannot ignore the events last week involving U.S. Secretary of War Pete Hegseth, his Department of War (DoW), and their declaration of war on leading AI developer Anthropic — events that are still unfolding. Mainstream media have so far treated the U.S. war on Iran as last week’s biggest news story, but I think they have made a mistake here, and that in terms of potential impact, the Hegseth/Anthropic thing is even bigger.
However, because of my earlier decision to prioritize skiing this week, I cannot comment at any length on this story, but I can compensate you, dear readers, with some excellent links:
Zvi Mowshowitz is reliable as always, and in his latest Substack post he offers a clear timeline of events along with extensive commentary.
Dean Ball, in his post Clawed, summarizes the essentials, and discusses the monumental importance of these events for the future of the United States as a republic.
Scott Alexander, with the help of some anonymous experts, explains the hairy but crucial details of what “all lawful use” means, in the context of OpenAI agreeing to replace Anthropic as the DoW’s main supplier of AI.
Readers who have noted my highly critical stance towards Anthropic in my recent posts Full speed ahead and Anthropic backpedals might be tempted to conclude that I would be against Anthropic and therefore pro-Hegseth on this matter. But that would be a mistake, because here I am fully on Anthropic’s side, for essentially the same reasons that Eliezer Yudkowsky explained as follows:
I'm not on your [DoW's] side, I'm not on Anthropic's side, I'm against whoever is currently behaving like more of a disaster. Usually that means I'm against the AI companies. But in this case, Hegseth is behaving like the villain in a disaster movie to an *even greater* extent than Dario Amodei's usual standard; and I am therefore, however locally and temporarily, on Anthropic's side of this one exact conflict.