Anthropic should stand its ground against the Pentagon

By Dave Lee

THEY SAY your values aren’t truly values until they cost you something. For Anthropic co-founder and Chief Executive Officer Dario Amodei, that cost might come as soon as Tuesday when he visits the Pentagon for what’s been billed as a showdown meeting with Secretary of Defense Pete Hegseth.

“This is not a friendly meeting,” a senior defense official . The AI company, best known for its chatbot Claude, took a more diplomatic tone: “We are having productive conversations, in good faith,” a spokesperson said.

The topic of discussion could not be more serious. To date, Claude has been the only AI model permitted to be used on the Pentagon’s classified networks, a valuable seal of approval that likely wins Anthropic enterprise contracts across many sectors but has also put it in a lonely position when pushing back against what it thinks might be some of the Pentagon’s intended use cases.

Hegseth is insisting Anthropic agree to sign on to allowing Claude to be used for “all lawful purposes,” threatening severe sanctions if it does not. That could mean not only the cancellation of a $200-million but also being designated a “supply chain risk,” a move that would force companies with military contracts to no longer use Anthropic’s AI for that work. The restriction would make Anthropic a less appealing partner for regular AI use cases, too.

Tensions between Anthropic and the Pentagon increased after the operation to capture Venezuelan President Nicolás Maduro, after which it was reported that Anthropic had objected to how its technology was used. The company has since said those reports were “inaccurate” and that no discussions were held nor complaints made. Even so, the Trump administration has branded Anthropic as the “woke” AI company for months. Amodei is “ideological,” Axios’ Pentagon source added. It’s important to look at the and consider whether these labels are justified.

The first issue Anthropic has is about mass surveillance. AI can be used to monitor masses of data and match up previously siloed datasets in ways that were previously impossible. The company worries that Fourth Amendment protections against unlawful surveillance don’t directly address what is possible with AI — leaving scope for actions it isn’t comfortable with from an administration that has demonstrated a willingness to test constitutional limits.

The second concern is about a technical practicality rather than an ethical red line. Anthropic has said it does not want its AI models to be used for control of autonomous weapons because it doesn’t believe its technology is reliable enough yet to make life-or-death decisions without human supervision.

If the Pentagon is unhappy with those apparently “woke” conditions, then, sure, it is well within its rights to cancel the contract. But to take the additional step of declaring Anthropic a “supply chain risk” appears unreasonably punitive while unnecessarily burdening other companies that have adopted Claude because of its superiority to other competing models.

Indeed, the quality of Anthropic’s product gives Amodei a stronger bargaining position than it might first appear. It would take a significant effort from the Pentagon to disentangle Claude within its systems and, even if it did, it would then need to find a willing and equally capable partner to fill the gap. Maybe that would be OpenAI or Google, but both would surely hesitate to lead their companies — and their employees — into the ethical quagmire Amodei is trying to avoid. Both companies are said to have been in talks with the Pentagon, but neither has reached a deal yet. The New York Times reported that Elon Musk had agreed to make his Grok chatbot available for use on classified material without the safeguards Anthropic was pushing for — though it is questionable whether Musk’s model will be as capable as Anthropic’s. 

From several angles, pressure is being applied on Anthropic to fall in line. In Tuesday’s meeting, Amodei must state it plainly: It is not “woke” to want to avoid accidentally killing innocent people. This isn’t a case of an arms maker dictating how the Pentagon must use a weapon it has purchased or against which target. No, this is a responsible company making sure a tool bought for one purpose won’t be recklessly used for another.

BLOOMBERG OPINION