Posts tagged: details

The Pentagon’s R&D chief said the Department of Defense was “scared” about Anthropic shutting off access to its AI during a critical moment.

During an appearance on the “All-In Podcast” posted on Friday, Undersecretary of Defense for Research and Engineering Emil Michael detailed two pivotal moments that culminated in the Pentagon formally designating Anthropic as a supply chain risk, effectively blacklisting one of the nation’s largest AI companies.

One of those instances, Michael said, was when Anthropic CEO Dario Amodei suggested that the impasse over how the Pentagon could deploy the AI startup’s models could be bridged with a phone call, even if it came during “a decisive moment.”

“I was giving these scenarios, these Golden Dome scenarios, and so on,” Michael said on “All-In Podcast,” describing President Donald Trump’s signature missile defense initiative.

“And he’s like, ‘Just call me if you need another exception.’ And I’m like, “But what if the balloon’s going up at that moment and it’s like a decisive action we have to take? I’m not going to call you to do something. It’s not rational.”

It’s not entirely clear what Anthropic would object to in the hypothetical Michael said he posed, though the implication is that some Golden Dome systems could have autonomous modes that fire weapons.

In the current US missile defense system, AI’s role is to provide rapid situational awareness and recommendations for human operators. AI could rapidly assess whether a detected launch poses a threat and recommend weapons to destroy it. Decisions on whether to listen to the recommendations are then made by air defense commanders.

Elsewhere in the interview, Michael said that part of the impasse with Anthropic is that he “can’t predict for the next 20 years what all the things we might use AI for.”

Michael, who was previously a top executive at Uber, said the department’s concerns about Anthropic began to escalate after the US conducted a targeted raid on Venezuela to capture Venezuelan President Nicolás Maduro. The assault raised major questions about sovereignty, and congressional democrats questioned the decision not to seek approval for the deployment of US forces.

In the wake of the raid, Michael said that an unnamed Anthropic executive called a Palantir executive to ask whether Anthropic’s AI models had been used to carry it out. The Pentagon accesses Anthropic’s AI models through a government cloud that is operated by Amazon Web Services and then run by Palantir, Michael said. (On February 27, President Donald Trump ordered federal agencies to stop using Anthropic’s AI, though that directive came with a six-month phase-out period.)

Michael said Palantir officials were so alarmed by Anthropic’s questions that they alerted him.

“I’m like, ‘Holy shit, what if this software went down, some guardrail kicked up, some refusal happened for the next fight like this one, and we left our people at risk,” Michael said, alluding to the US’s current war against Iran.

As talks grew heated, Michael said he felt like Anthropic turned the discussion “into a PR game” by publicly raising concerns about how the terms the Pentagon sought would not adequately account for potential misuse. Amodei has confirmed that Anthropic was particularly worried about the risks posed by fully autonomous weapons and how powerful AI models could be abused to spy on American citizens.

During the heated back-and-forth, Michael publicly called Amodei a “liar” with “a God-complex.”

On Thursday, the Pentagon said it formally notified Anthropic that it was declaring the company and its products to be a supply chain risk, the first time in history that label had been applied to a US company.

Amodei responded that his AI startup had “productive conversations” with the Pentagon in recent days, but Michael later said that no discussions were ongoing.

Anthropic has suggested it will challenge the designation in court, especially since Defense Secretary Pete Hegseth has said it prevents any defense contractor from doing business with Anthropic.

Asked about why the Pentagon went so far, Michael said the designation was not “punitive.”

“If their model has this policy bias, let’s call it, based on their constitution, their culture, their people, and so on,” he said. “I don’t want Lockheed Martin using their model to design weapons for me.”

Earlier this week, a Lockheed spokesperson said it would follow Trump and the Pentagon’s direction on whether it would continue to use Anthropic’s products. Michael also called out Boeing, describing how the airplane manufacturer could use Anthropic’s AI for non-defense tasks.

“So, Boeing wants to use Anthropic to build commercial jets — have at it,” he said. “Boeing wants to use it to build fighter jets. I can’t have that because I don’t trust what the outputs may be, because they’re so wedded to their own policy preferences.”

While Michael was critical of Anthropic, he praised xAI and Elon Musk for agreeing to the department’s terms, allowing it to deploy AI “for all lawful uses.”

Michael also praised OpenAI and its CEO, Sam Altman, for working with the Pentagon to quickly stand up another AI system capable of operating in classified settings, so the department can phase out Anthropic.

Altman and OpenAI have received significant blowback online for agreeing to work with the Pentagon. Altman publicly urged the department not to label Anthropic a supply chain risk.

“To his credit, I called him and said, ‘I need a solution if this thing goes sideways. I need multiple solutions. I’d like you to be one of them,” Michael said. “And he’s like, ‘Okay, well, what can I do for the country?’ I was like, ‘I need to get you up running as soon as I can.'”