Anthropic Under Scrutiny for Dropping Safety Pledge Amid Pentagon Pressure
Anthropic has reportedly dropped its flagship safety pledge, a move that comes as the company faces pressure from the U.S. military to loosen AI guardrails for defense applications. The Pentagon has allegedly threatened to exclude Anthropic from defense contracts if it does not permit more flexible military use of its models. The company is said to be in a standoff with officials over the issue.
- The now-abandoned "Responsible Scaling Policy" (RSP) was a framework first published in September 2023 to manage catastrophic AI risks by tying model capabilities to specific safety and security standards. The policy established "AI Safety Levels" (ASLs), similar to biosafety levels, that required stricter controls as a model's potential for harm increased, with a key pledge to halt scaling if safety measures couldn't keep pace. - Anthropic's specific "red lines" that have caused the dispute with the Pentagon are prohibitions on using its models for autonomous weapons targeting and domestic mass surveillance. CEO Dario Amodei has publicly warned that an AI analyzing billions of conversations could be used to suppress dissent. - The Pentagon's ultimatum to Anthropic includes threatening to designate the company a "supply chain risk," which would prevent any company with a government contract from using Anthropic's software. Officials have also mentioned invoking the Defense Production Act to compel Anthropic to comply with military requirements. - This conflict is unfolding in the context of the Pentagon's new "AI Acceleration Strategy," released in January 2026, which mandates an "AI-first warfighting force". The strategy requires that commercial AI models be deployable for "any lawful use" within 30 days of their public release and free from "ideological 'tuning'". - While Anthropic's Claude was the first and, until recently, the only frontier model deployed on classified Pentagon networks, competitors are signaling more willingness to work with the military. OpenAI and Elon Musk's xAI have reportedly agreed to the Pentagon's "all lawful purposes" standard, with xAI's Grok now being deployed on the same classified networks as Claude. - Anthropic updated its RSP on February 24, 2026, removing the binding commitment to pause development if safety couldn't be guaranteed. The company stated the change was due to the rapid advancement of AI by competitors and the slow pace of government action on AI safety, though it maintains the timing is unrelated to the Pentagon dispute. - The Pentagon's Chief Technology Officer, Emil Michael, has publicly stated it is "not democratic" for a single company to dictate policy on military AI use beyond what is determined by Congress and the President. In July 2025, the DoD had awarded contracts of up to $200 million each to Anthropic, OpenAI, Google, and xAI to customize their models for military applications. - The immediate trigger for the confrontation appears to be the suspected use of Claude, via a partnership with Palantir, in a January 3rd U.S. military operation in Venezuela that led to the capture of Nicolás Maduro. Following this, CEO Dario Amodei reiterated the company's "bright red lines" to the Department of Defense.
