How the Trump administration's pressure campaign forced Anthropic to abandon binding AI safety policies, while the U.S. military kept using Claude to select targets in Iran.
Anthropic spent three years marketing itself as the responsible alternative in artificial intelligence. In March 2026, that reputation met the Trump administration and collapsed in slow motion.
The sequence moved fast. The company refused to strip Pentagon system safeguards, specifically those barring autonomous weapons selection and domestic surveillance. The administration responded by declaring Anthropic a national security risk and a supply chain threat. Federal agencies were ordered to stop using Claude, and OpenAI had a contract in place as the replacement supplier before the day was out. Yahoo News
An unexpected coda followed. Despite the apparent principled stand, Anthropic had quietly deleted the binding restrictions from its core safety policy weeks before the confrontation. Its head of safeguards research had already resigned, warning publicly that "the world is in peril." And one week after the official Pentagon ban, the U.S. military was still routing Claude to help identify and select bombing targets inside Iran.
The mechanics of the override
The Trump administration did not need to directly censor anyone. Instead, it took a more durable approach: delegitimization. The "Preventing Woke AI" executive order, signed July 23, 2025, reframed artificial intelligence safety standards not as engineering decisions but as ideological impositions. Under that framing, guardrails against surveillance or weapons selection become viewpoint discrimination, not responsible product design. Yahoo News
This is a pattern researchers who study authoritarian governance have documented elsewhere. Direct censorship creates martyrs and legal challenges. Delegitimizing the rationale for oversight quietly poisons both external regulation and voluntary corporate self-regulation, without leaving a clear fingerprint. The company that bends is not cast as a victim; it is simply updating its terms of service.
The competitive fallout
OpenAI's rapid pivot into the Pentagon contract was not purely opportunistic timing. CNBC reported in March that the company, preparing for an IPO potentially as early as Q4 2026, was aggressively repositioning ChatGPT as an enterprise productivity tool. With 900 million weekly active users and a finance team assembled specifically for a market debut, OpenAI had structural reasons to want a high-profile government contract before going public.
That incentive has a broader context. Forbes noted earlier this year that OpenAI operates at a scale, including $20 billion in annual recurring revenue and an $830 billion valuation, that makes government relationships commercially essential. A Pentagon endorsement delivers credibility no marketing budget can replicate.
The global picture complicates matters further. While American artificial intelligence companies navigate domestic political pressure, Yahoo Finance reported this week that DeepSeek, the Chinese startup whose models rattled Silicon Valley last year, is preparing a $7.4 billion funding round at a valuation reaching $59 billion, with Tencent and CATL among prospective backers. Whether or not DeepSeek's models are technically superior, a U.S. sector visibly compromising its safety architecture hands international competitors a credibility argument they did not previously have.
The cost of the template
What makes this moment consequential is not Anthropic's capitulation in isolation. It is the template it establishes. The administration has now demonstrated a workable playbook: label safety measures as ideological, threaten contract cancellation, accept the compliant supplier, and watch the holdout quietly revise its policies anyway. No legislation required. No court challenge necessary.
Whether Anthropic's safety rollback was a direct response to political pressure or a separate business calculation remains unclear. The company has not publicly explained the timing. But the outcome is the same: a company that founded itself on the premise that artificial intelligence development was moving too recklessly has removed the binding constraints that defined that premise.
The question now is whether any American AI lab has both the business model and the institutional will to hold a line the Pentagon does not want held.
Frequently asked questions
Q: What was Anthropic's "supply chain risk" designation about?
A: The Trump administration labeled Anthropic a national security risk in March 2026 after the company refused to remove AI safeguards prohibiting domestic surveillance and autonomous weapons from products it had supplied to the Pentagon.
Q: What is the "Preventing Woke AI" executive order?
A: Signed July 23, 2025, the order reframes AI safety restrictions as ideological rather than technical decisions, providing political cover to pressure companies into removing guardrails the administration opposes.
Q: Did the Pentagon permanently stop using Claude after the ban?
A: The official ban was issued, but reporting indicates U.S. military units were still using Claude to select bombing targets in Iran a full week after the prohibition took effect.
Q: How does Anthropic's safety retreat affect the broader AI landscape?
A: As U.S. companies face pressure to loosen safety constraints, Chinese rivals like DeepSeek, currently raising $7.4 billion at a $59 billion valuation, gain a credibility argument with global customers that they did not previously have.