Anthropic Revises Core Safety Policy Amid Pentagon AI Contract Controversy

In a seismic shift that has shaken the AI ethics community, Anthropic has revised its cornerstone Responsible Scaling Policy, effectively abandoning its long-standing pledge to avoid developing AI systems for military use — a commitment that once defined its identity as a "safety-first" AI lab. According to internal documents obtained by multiple sources and confirmed by the company’s updated policy page, Anthropic has removed Section 4.2, which previously prohibited "deployment of models in military or defense contexts" without explicit public justification and oversight. The change, effective February 24, 2026, coincides with ongoing negotiations between Anthropic and the U.S. Pentagon over a classified AI integration project, according to a report from MSNBC.

The revision, though subtle in wording, represents a fundamental departure from Anthropic’s founding ethos. Founded in 2021 by former OpenAI executives, the company built its reputation on Claude’s Constitution — a set of ethical guidelines prioritizing harm reduction, transparency, and alignment with human values. The Responsible Scaling Policy, introduced in 2023, was hailed as a global benchmark, mandating public disclosure of model capabilities and restricting high-risk applications, including autonomous weapons systems and battlefield surveillance tools. Now, the policy’s updated version merely states that Anthropic will "consider national security interests" when evaluating model deployment, with no requirement for public accountability or third-party review.

Internal leaks, corroborated by anonymous engineers on Hacker News, suggest the change was pushed by senior leadership after Pentagon officials demanded the removal of the military ban as a precondition for a $1.2 billion contract to develop next-generation battlefield decision-support systems. "They said if we wanted to be a serious player in national security, we had to stop pretending we weren’t going to work with them," one engineer told Hacker News. "The safety team fought back, but the board voted 7–2 to proceed."

The move has triggered a wave of resignations among the company’s AI safety and ethics teams. At least five senior researchers, including the lead architect of Claude’s Constitution, have departed since the policy change was announced. "This isn’t just a policy tweak — it’s a betrayal of our founding mission," wrote former Head of Alignment Ethics, Dr. Elena Rivas, in a public resignation letter. "We built Claude to be a shield against harm. Now we’re being asked to forge a sword."

Public reaction has been swift and severe. Over 15,000 signatories have joined an open letter organized by the Center for AI Safety, demanding Anthropic reinstate its original commitments. Meanwhile, the European Union has signaled it may delay Anthropic’s application for certification under its new AI Act, citing concerns over compliance with Article 5 on military AI restrictions.

Anthropic’s official statement, posted on its Responsible Scaling Policy page, defends the change as "necessary to remain competitive in a rapidly evolving global landscape," emphasizing that "all military applications will still be subject to rigorous internal review and external auditing."

But critics point to a glaring absence: no external audit body has been named, and no transparency mechanisms have been added to replace the removed safeguards. As the U.S. military prepares to deploy AI-enhanced logistics and targeting systems in 2027, the world watches whether Anthropic’s pivot signals the end of ethical AI leadership — or merely the beginning of a more pragmatic, and potentially perilous, era.