The Big Problem with Anthropic’s ‘AI Safety’ Brand
The Big Problem with Anthropic’s ‘AI Safety’ Brand
The company’s current battle with the Pentagon distracts from the military use of its flagship model.
The San Francisco–based AI company Anthropic has garnered national attention after a high-profile public dispute with the Pentagon over AI safety standards and the potential use of its technology for mass domestic surveillance. Its defiance of demands from the Trump administration—which last month led to a ban on federal government contracting with the company that is now the basis of an Anthropic lawsuit against the administration—prompted Silicon Valley elites to rally behind Anthropic and its CEO, Dario Amodei.
The episode has rallied Silicon Valley behind the company in ways reminiscent of the tech world’s pre-Trump liberal era. “Now what began as a whisper of support for Anthropic in the tech industry has crescendoed into a shout,” the New York Times reports.
The independent journalist Jack Poulson told The American Conservative he suspects that winning such support may be Anthropic’s goal—that the feud with the Trump administration may function less as a principled stand for civil liberties than as a calculated marketing strategy. That strategy would be designed to appeal to Silicon Valley liberals and progressives who dislike Trump, are skeptical of how the government uses their technology, yet are ultimately willing to provide the U.S. and other governments with the most powerful tools to censor, surveill, and even kill their enemies for the right price.
Poulson, who left his role as a senior scientist at Google in 2017 in protest over the company’s work on a censored search engine project for China, pointed to the fact that Anthropic’s entire brand distinction from OpenAI rests on its supposed ethical standards. That Anthropic is now engaged in a highly public dispute—one that has reportedly helped propel its Claude chatbot past ChatGPT in March app downloads—is “on brand,” Poulson said. It may simply be a way “for Anthropic to establish itself as #resistance,” he argued, so that “its employees can still feel welcome in liberal circles.”
Anthropic co-founder Jack Clark, for his part, has experience with such calculated marketing strategies from his time as OpenAI’s policy director, helping that company exploit the false pretense of being a nonprofit for financial benefits.
Poulson pointed to a series of under-discussed disclosures that call into question Anthropic’s image as a company uniquely guided by ethical concerns about surveillance and the misuse of artificial intelligence. A leaked meeting booklet uncovered by Poulson in 2023 reveals Anthropic representatives participating in a closed-door intelligence collaboration involving senior CIA officials—including the agency’s chief technology officer and director of artificial intelligence—alongside Australian government officials.
The workshop, organized through forums run by former Google CEO Eric Schmidt’s Special Competitive Studies Project and the Australian Strategic Policy Institute, was part of a broader effort to explore how large language models could be integrated with Western security states. At the same time, Anthropic has expanded its government business, hiring longtime Palantir employee Steve Sloss to lead U.S. government sales and pitching its technology to various intelligence agencies, including the National Geospatial-Intelligence Agency.
Those disclosures raise broader questions about Anthropic’s ties to the deep state. Poulson noted that the company has partnered with Palantir, whose platforms are built around large-scale commercial and government data fusion, despite Anthropic’s public warnings about the risks of such systems. He also pointed to the CIA’s Open Source Enterprise, which has long discussed using large language models to process vast troves of publicly available data. The 2023 meeting between Anthropic and CIA officials raises questions about Anthropic’s awareness of, or cooperation with, those efforts.
Even as the firm now publicly resists certain Pentagon demands, its flagship AI model, Claude, has already been integrated into the U.S. military’s targeting infrastructure where it operates under the hood of Palantir’s Maven system. The model is now used by military planners to analyze intelligence feeds and generate prioritized target lists for U.S. strikes in Iran, and was likely involved in the American bombing of an elementary school which killed more than 160 people, mostly young girls. Previously, Anthropic’s AI was reported to have played a role in the operation to kidnap Venezuela’s leader Nicolas Maduro in which more than 80 people were killed.
Though it may advertise itself as opposed to fully autonomous weapons, Anthropic seems to have few red lines about how its products are used to kill non-Americans.
While Anthropic is now widely celebrated as a champion of civil liberties and AI ethics, its cozy relationship with the U.S. national security state—and the growing use of its technology in military operations—should invite far more skepticism about the company’s branding than it has so far received.
The post The Big Problem with Anthropic’s ‘AI Safety’ Brand appeared first on The American Conservative.