Anthropic Defies Pentagon Demands to Loosen AI Safety Safeguards

The escalating tension between Anthropic and the United States Department of Defense represents a critical stress test for the "AI safety" movement. For years, Anthropic has marketed itself as a Public Benefit Corporation, prioritizing the ethical alignment of its Claude models over rapid commercial expansion. However, the reported involvement of Claude in the January abduction of Venezuelan President Nicholas Maduro has forced a collision between these corporate values and the strategic objectives of the Trump administration. Defense Secretary Pete Hegseth’s ultimatum—demanding that Anthropic loosen its usage restrictions by Friday or lose its government contracts—signals a new era where the Pentagon views private-sector safety guardrails as obstacles to national security.

The specific points of contention revolve around two of the most sensitive areas in AI ethics: domestic surveillance and lethal autonomous weapons systems (LAWS). Anthropic’s current terms of service strictly prohibit the use of its technology for monitoring US citizens or for programming weapons that can select and engage targets without human intervention. The Pentagon, however, appears to be seeking a broader mandate, arguing that these restrictions limit the effectiveness of AI in modern warfare and intelligence gathering. This demand places Anthropic in a precarious position, as its identity is built on the premise of "Constitutional AI," a method of training models to follow a specific set of ethical principles.

“

For years, Anthropic has marketed itself as a Public Benefit Corporation, prioritizing the ethical alignment of its Claude models over rapid commercial expansion.

From a market perspective, this rift could reshape the competitive landscape of the defense tech sector. Anthropic was the first AI developer to be integrated into classified Pentagon operations, a major win that validated its technical prowess. If the company is forced out, it creates a vacuum that competitors like OpenAI or specialized defense contractors like Palantir and Anduril may be eager to fill. While OpenAI has also historically maintained safety guidelines, the pressure to align with the administration’s "America First" AI policy may lead to more flexible interpretations of those rules.