Did Dario Amodei Ask Claude If He Was Being An Idiot?
Life comes at you fast in the Trump era. Less than 24 hours ago, the bombshell story was Trump/Hegseth dropping the sledgehammer on Anthropic. Then I wake up to the news of US-Israeli decapitation strikes on Iran.
I’ll wait a bit for the dust to settle (literally and figuratively) before reacting to today’s news. Yesterday’s news deserves comment.
In brief, Anthropic–the company that created Claude AI–was in negotiations with the Department of War regarding use of customized versions of Claude on classified DoW systems. The negotiations stalled over the DoW insistence that Anthropic license Claude for all “lawful uses.” Anthropic insisted that the Pentagon not use it for “mass surveillance” or “fully autonomous weapons.”
An hour before the deadline for an agreement, Secretary of War Hegseth announced that the negotiations had been terminated. But that was only the beginning.
After Hegseth made that announcement, I mused on X that Anthropic was playing a dangerous game:
I was thinking of surreptitious pressure, with Bessent and others telling Anthropic’s input and capital suppliers to think twice about dealing with the company. But no. Within hours of announcing the end of the negotiations, Hegseth designated the company as a Supply Chain Risk, and forbade any company doing business with the Department of War from dealing with Anthropic.
Bad on me for thinking the administration would take the subtle approach!
This is a death sentence.
No doubt Anthropic will sue, and will move for an injunction. They may even get one. But that will not really help.
Consider all the suppliers to and collaborators with Anthropic (which includes Google, Amazon, and others). The risk of dealing with Anthropic just skyrocketed. Specifically, the legal risk. While litigation plods forward–and believe me, I can tell you from extensive experience Federal litigation plods–all these firms will calculate that any long-term engagement with the company could get blown up if the courts eventually side with the USG. So long term contracts, joint ventures, equity investments, and loans have a huge downside risk that didn’t exist 24 hours ago.
Real options analysis shows that the rational response to a spike in uncertainty is to defer making decisions that lock you in for an extended period until the uncertainty is resolved. So even if an injunction allows companies to continue to deal with Anthropic, they will rationally back off–way off–pending resolution. Yeah, maybe paper clip suppliers will continue to sell to Anthropic, but the real strategically important players won’t want to get locked into long-term arrangements with it.
Competition in AI is intense. The speed of innovation is intense. The legal risk itself puts Anthropic at a severe competitive disadvantage in such a hyper-fast-paced competitive environment. Thus, the punishment meted out by the process will likely be just as fatal as the actual implementation of the DoW order.
Some are calling this corporate murder. It isn’t. It is corporate suicide.
Seriously. Didn’t Anthropic CEO Dario Amoedi and his board ask Claude to generate scenarios of the effect of its refusal to toe the DoW line and which of those scenarios would be the most likely? Did he ask it if he was being an idiot? Because he was.
Now to the rights and wrongs. Here is Anthropic’s justification for its obduracy:
We support the use of AI for lawful foreign intelligence and counterintelligence missions. But using these systems for mass domestic surveillance is incompatible with democratic values. AI-driven mass surveillance presents serious, novel risks to our fundamental liberties. To the extent that such surveillance is currently legal, this is only because the law has not yet caught up with the rapidly growing capabilities of AI. (Emphasis added).
In other words, Amodei wants to arrogate to himself the power to require the Pentagon to act as HE thinks SHOULD BE legal, not what is legal. He wants to usurp the power of all three branches of the Federal government. That is obviously not his job.
If people believe the Pentagon has violated the law and the Constitution (in interviews he speaks of Fourth Amendment violations), they can bring suit. It is up to the courts, not Dario Amodei, to decide whether the Pentagon has or has not done so. Furthermore, Congress can act to “catch up” the law with the capabilities of AI. It’s not Dario Amodei’s job to become a law of his own.
Amodei’s argument is self-contradictory. On the one hand, he says he supports “lawful” uses. But then he says, not really, because he doesn’t like the law. Pick one, dude.
Now, you might counter that OK, then the US should just refuse to deal with Anthropic, but not try to whack it.
The Pentagon has clearly identified Claude as vital to the national defense. And, ironically, the company’s claims about its capabilities make the Pentagon’s case for it. So, the US needs Claude. The company refuses to provide it on acceptable terms. So it is necessary to compel them.
Or maybe you think Anthropic’s conditions are reasonable. They are not. These terms would essentially allow the company, at its whim, to impair the DoW’s use of Claude. That is, they could “hold up” the Pentagon.
Claude is a very specific asset, especially since it would be customized for Pentagon use. Even if DoW also employed other AI, it would not be able to replace at short notice the bespoke capabilities of Claude.
No rational party will subject themselves to such a massive holdup risk. The Pentagon clearly is rational about this.
Further, Anthropic would have legal recourse in the event it believed that the Pentagon was using Claude unlawfully.
Anthropic lamely argues that
They have threatened to remove us from their systems if we maintain these safeguards; they have also threatened to designate us a “supply chain risk”—a label reserved for US adversaries, never before applied to an American company—and to invoke the Defense Production Act to force the safeguards’ removal. These latter two threats are inherently contradictory: one labels us a security risk; the other labels Claude as essential to national security.
No contradiction at all. The “safeguards”–basically, giving Anthropic a kill switch that it can activate at is whim–are utterly incompatible with Pentagon use. Claude is a vital national security tool. Hence, force it is. Perfectly consistent logical sequence of actions.
It is ironic that Anthropic is not willing to accede to the Pentagon mere weeks after it screamed about three Chinese firms training their AI systems on Claude, thereby allowing the copycats to create systems with a substantial fraction of Claude’s capability. Meaning that the Chinese already likely have the capabilities that fret Amodei so much. And you can be totally sure that China will respect Amodei’s tender concerns for our liberties, amirite? But the US government is not to be trusted with these same capabilities, apparently.
A couple of admittedly subjective reactions.
When I first looked at some Amodei vidoes, I got a Sam Bankman Friedman vibe.
And wouldn’t you know, like felon Fried, Amodei is an “effective altruism” guy. This is disturbing. EA is a hyper-utilitarian “philosophy” that appeals to would be ubermenschen who often use it to justify their own psychopathy. SBF being the (imprisoned) poster child for this.
And as a general matter, be very, very, very suspicious of companies (and CEOs) that set themselves out as moral paragons. Remember “Don’t be evil”? Yeah, good times, good times. Or Bill Gates? I could go on. (And yes, Elon Musk fits too).
Lastly, maybe Amodei is suffering from Oppenheimer-like guilt and remorse over creating a power that has escaped his control. His attempt to restrict Pentagon use of the genie he freed from the bottle could be a desperate attempt to re-exert control. But he can’t. The genie is well and truly gone, and this faceoff with the Pentagon will in no way stop it.
Categories: Economics, Military, Politics