Pentagon-Anthropic Row: Will Startups Avoid Defense?

In just over a week, a series of rapid developments unfolded: negotiations over the Pentagon’s utilization of Anthropic’s Claude technology collapsed, the Trump administration officially designated Anthropic a supply-chain risk, and the AI company promptly announced its intention to legally challenge this classification in court.

Meanwhile, OpenAI swiftly unveiled a deal of its own, a move that triggered considerable public backlash. This reaction manifested in users uninstalling ChatGPT and propelled Anthropic’s Claude to the top of App Store charts. Adding to the internal dissent, at least one OpenAI executive reportedly resigned, citing concerns that the announcement was rushed without adequate safeguards in place.

On a recent episode of TechCrunch’s Equity podcast, Kirsten Korosec, Sean O’Kane, and I delved into the broader implications of these events for other startups aspiring to collaborate with the federal government, especially the Pentagon. Korosec specifically questioned, "Are we going to see a changing of the tune a little bit?"

Sean O’Kane highlighted the unusual nature of this situation for several reasons. Primarily, he noted that both OpenAI and Claude produce technologies "no one can shut up about," ensuring constant public discourse. Crucially, he added, the core of this dispute centers on "how their technologies are being used or not being used to kill people," which naturally intensifies scrutiny.

Despite these unique factors, Kirsten Korosec maintained that the entire scenario should "give any startup pause."

Kirsten elaborated on her initial query, pondering whether other startups might now reconsider pursuing federal funding, particularly after witnessing the contentious "debate and wrestling match" between the Pentagon and Anthropic. She reiterated her concern: "Are we going to see a changing of the tune a little bit?"

Sean shared his skepticism, suggesting that in the near term, the answer might largely be "no." He reasoned that numerous companies, from nascent startups to established Fortune 500s, routinely engage in work with the government, specifically the Department of Defense or the Pentagon, and for many, "that work flies under the radar."

He cited General Motors as an example, noting its long-standing history of manufacturing defense vehicles for the Army, including recent electric and autonomous versions, work that "never really hits the zeitgeist." The fundamental problem for OpenAI and Anthropic, he argued, is that their products are widely used and, more significantly, are ones "no one can shut up about."

This intense public spotlight, he explained, naturally magnifies their involvement to a degree that most other federal government contractors, particularly those contributing to war-fighting elements, typically do not encounter.

Sean added a critical caveat: much of the fervor surrounding the discussions between Anthropic, OpenAI, and the Pentagon is explicitly focused on "how their technologies are being used or not being used to kill people, or in parts of the missions that are killing people." He emphasized that this goes beyond mere public attention or brand familiarity, introducing an "extra element" that feels more abstract when considering a traditional defense contractor like General Motors.

Consequently, he doesn't anticipate companies like Applied Intuition, which position themselves as "dual use," significantly withdrawing from such engagements. This is primarily because they do not attract the same level of public scrutiny, and there isn't a comparable "shared understanding of what that impact might be."

I interjected, suggesting that this particular narrative is profoundly "unique and specific to these companies and personalities in a lot of ways." While acknowledging the value of broader "thought pieces" exploring the role of technology, especially AI, in government, I underscored the distinctiveness of this specific case.

I also found it a "very curious lens" through which to examine these issues, noting that Anthropic and OpenAI, despite the public drama, are not fundamentally different in many of their stated positions. It's not a scenario where one company rejects government work while the other embraces it, or one permits unrestricted use while the other demands limitations. Both, at least publicly, assert, "We want restrictions on how our AI gets used." The primary distinction, I observed, appeared to be Anthropic's more resolute stance, "digging in their heels a lot more about: You cannot change the terms in this way."

Adding another layer of complexity, I highlighted a possible "personality layer," citing reports that the CEO of Anthropic and Emil Michael, currently the Chief Technology Officer for the Department of Defense (and recognizable to many TechCrunch readers from his Uber tenure), reportedly "just really don’t like each other."

Sean concurred, adding, "Yes, there’s a very big 'girls are fighting' element here that we should not overlook."

Kirsten acknowledged the personal dynamic but quickly pivoted, asserting that "the implications are a little bit stronger than that." She reiterated the core dispute between the Pentagon and Anthropic, noting that Anthropic "appears to have lost," despite its technology still being "very much being used by the military" and considered "a crucial technology." She also pointed out that OpenAI had "kind of stepped in," and the situation remained fluid, likely to evolve further.

The subsequent public "blowback" for OpenAI has been noteworthy, marked by a significant surge in ChatGPT uninstalls, which she estimated "surged 295% after OpenAI locked in the deal with the Department of Defense."

For Kirsten, however, much of this constituted "noise" distracting from the "really critical and dangerous thing": the Pentagon's attempt to unilaterally alter "existing terms on an existing contract." She stressed that this is "really important and should give any startup pause," as the current "political machine" within the DoD seems to operate differently, diverging from normal procedures where government contracts typically "take forever to get baked in." The very act of seeking to change established terms, she concluded, presents a significant problem.