AI Safety and the Military Industrial Complex

When Anthropic was cleared by the US government last year for classified use—including military applications—it didn’t generate much buzz. However, this week, a significant development emerged: The Pentagon is reevaluating its partnership with the company, which includes a $200 million contract. This reconsideration seems to stem from Anthropic’s reluctance to take part in certain lethal operations. The Pentagon might classify Anthropic as a “supply chain risk,” a designation typically reserved for entities dealing with countries under federal scrutiny, like China, potentially barring the use of Anthropic’s AI by defense contractors. Chief Pentagon spokesperson Sean Parnell acknowledged the scrutiny, stating, “Our nation requires that our partners be willing to help our warfighters win in any fight. Ultimately, this is about our troops and the safety of the American people.” This sends a clear warning to other companies: OpenAI, xAI, and Google, currently holding Department of Defense contracts for unclassified work, are now racing to secure their own high-level clearances.

There’s a lot to unpack here. A key issue is whether Anthropic is facing consequences for raising concerns about its AI model, Claude, being used in the operation to oust Venezuela’s president Nicolás Maduro (an assertion the company disputes). There’s also their public advocacy for AI regulation—a stance that diverges sharply from the prevailing industry norms and the administration’s approach. Yet, the deeper question is whether government military demands could compromise AI safety.

Many in the field view AI as the most transformative technology ever created. The majority of current AI companies were founded on the belief that achieving AGI, or superintelligence, can be done in a manner that avoids significant harm. Elon Musk, founder of xAI, was once a leading voice advocating for AI restraint, co-founding OpenAI out of fear for the technology being too dangerous in the hands of profit-oriented organizations.

Anthropic stands out as the most safety-focused of these companies. Their mission centers on embedding robust guardrails within their models to prevent bad actors from exploiting AI’s potential for harm. Isaac Asimov articulated this best in his laws of robotics: A robot may not injure a human being or, through inaction, allow a human being to come to harm. Even when AI surpasses human intelligence—a scenario that many in AI leadership anticipate—these guardrails need to remain intact.

It feels paradoxical that prominent AI labs are eager to integrate their technologies into advanced military and intelligence operations. As the first major lab given a classified contract, Anthropic offers the government a “custom set of Claude Gov models built exclusively for U.S. national security customers.” However, Anthropic insists it has adhered to its safety protocols, including a ban on using Claude for weapon development. CEO Dario Amodei has explicitly stated his opposition to involving Claude in autonomous weaponry or AI surveillance by the government. Yet, the current administration may not align with this view. Department of Defense CTO Emil Michael, previously Uber’s chief business officer, remarked that the government won’t accept restrictions from an AI firm on how the military can deploy AI in weaponry. “If there’s a drone swarm coming out of a military base, what are your options to take it down? If the human reaction time is not fast enough … how are you going to?” he posed, highlighting a disregard for the first law of robotics.

One could argue that effective national security necessitates access to cutting-edge technology from the most innovative firms. While tech companies hesitated to collaborate with the Pentagon just a few years ago, by 2026, they have largely embraced military contracting. Although I have yet to hear any AI executive connect their models with lethal applications, Palantir CEO Alex Karp is open about stating, with apparent pride, “Our product is used on occasion to kill people.”