--- title: Pentagon Threatens to Blacklist Anthropic Over AI Guardrails description: The Pentagon may designate Anthropic a supply chain risk if it refuses to lift safety restrictions on military use of Claude. author: Darie Nani (Editor-in-Chief) date: 2026-02-24T14:03:16.000Z updated: 2026-04-22T12:25:40.049Z canonical: https://www.sovereignmagazine.com/article/pentagon-threatens-to-blacklist-anthropic-over-ai-guardrails image: https://cdn.nanimediahouse.com/pentagon-hegseth-anthropic-ai-guardrails.webp categories: Artificial Intelligence, Politics content_type: News region: United States publication: Sovereign Magazine about: - type: Organization name: Anthropic --- Defence Secretary Pete Hegseth will meet Anthropic CEO Dario Amodei at the Pentagon on Tuesday. A defence official told [Axios](https://www.axios.com/2026/02/23/hegseth-dario-pentagon-meeting-antrhopic-claude) that Hegseth would present Amodei with an ultimatum: remove all company-imposed safety restrictions on military use of Claude, or face designation as a ‘[supply chain risk](https://www.sovereignmagazine.com/article/judge-blocks-pentagon-anthropic-ban)’. That label, normally reserved for foreign adversaries, would require every [defence contractor](https://www.sovereignmagazine.com/article/how-a-seven-year-old-startup-just-landed-a-1-1b-defense-contract-that-could-reshape-military-) to certify that it does not use Anthropic products. The effect would extend far beyond the Pentagon’s own $200 million contract with the company, signed in July 2025. It would shut Claude out of the entire US defence-industrial base. ## Anthropic’s stated limits Anthropic has two red lines. It will not permit Claude to be used for mass surveillance of Americans or for fully autonomous weapons systems. The company is willing to negotiate on other terms but has refused the Pentagon’s demand that it make Claude available for ‘all lawful purposes’ without restriction. The dispute escalated in January after reports that Claude was used during the military operation to capture Venezuelan President Nicolás Maduro. Anthropic said it had not identified any policy violations from the reported use, but the incident drew scrutiny to the gap between the company’s stated limits and operational reality. ‘We should arm democracies with AI, but we should do so carefully,’ Amodei said. ## The Pentagon’s position [Hegseth](https://www.sovereignmagazine.com/article/pete-hegseth-nickname-dumb-mcnamara-pentagon)’s January 2026 AI strategy document requires all military AI contracts to eliminate company-specific guardrails within 180 days. Pentagon spokesman Sean Parnell said: ‘Our nation requires that our partners be willing to help our warfighters win in any fight.’ Undersecretary of Defence Emil Michael confirmed that negotiations stalled over disagreements on permitted use cases. The practical concern is Claude’s position on classified networks. Anthropic, through its partnership with [Palantir](https://www.sovereignmagazine.com/article/palantir-nhs-contract-karp-manifesto), is the only AI company currently deployed on the systems used for the Pentagon’s most sensitive [intelligence operations](https://www.sovereignmagazine.com/article/advanced-geospatial-analysis-tools-herald-new-era-for-intelligence-operations). An LLM with a large context window can process bulk intelligence data (intercepted communications, metadata, financial records) at a speed and scale that individual analysts cannot match. That capability is precisely what Anthropic wants contractual limits on. ## xAI already signed The same week Hegseth summoned Amodei, Elon Musk’s xAI signed a deal to deploy Grok on classified military systems. xAI agreed to the ‘all lawful use’ standard without conditions. Google and OpenAI are also in talks to move their models into the classified space. The Pentagon does not need Anthropic. It has alternatives ready. Anthropic’s $200 million contract is small against the company’s $14 billion annual revenue and $380 billion valuation after its February Series G round. But the supply chain risk designation would not just end a single contract. It would force Lockheed Martin, Raytheon, Palantir, Booz Allen and every other defence vendor to choose between Anthropic and the Pentagon. That is the actual leverage. ## Further Context **Q: What do guardrails mean in AI?** AI guardrails are policies, technical controls and contractual terms that restrict how an AI system can be used. In commercial products, guardrails typically prevent the model from generating harmful content or assisting with illegal activity. In military contexts, the term refers to use-case restrictions imposed by the AI vendor (such as prohibiting autonomous targeting or bulk surveillance) rather than content filters. The Pentagon’s current dispute with Anthropic centres on whether the vendor or the customer sets those boundaries. **Q: Did Google drop the ban on AI weapons?** Google reversed its internal prohibition on AI for weapons and surveillance in February 2025. The original ban dated to 2018, when thousands of Google employees protested Project Maven, a Pentagon programme that used AI to analyse drone footage. Amnesty International described the reversal as enabling technologies including mass surveillance, semi-automated drone strikes and AI-generated target lists. Google, OpenAI and xAI have all since agreed to make their models available to the US military on the Pentagon’s terms. **Q: Are there fully autonomous weapons?** No weapon system currently in deployment selects and engages human targets without a human operator in the decision chain. The US Department of Defense Directive 3000.09 requires human involvement in lethal force decisions. However, systems like Israel’s Harpy loitering munition and AI-assisted target generation tools (such as those reported in Gaza) operate with diminishing levels of human oversight. The concern driving Anthropic’s red line is that an LLM integrated into targeting workflows could reduce the human role to a procedural rubber stamp. **Q: What is the law on autonomous weapon systems?** There is no binding international treaty on autonomous weapons. The UN Convention on Certain Conventional Weapons has hosted discussions since 2014 without producing a legal instrument. In November 2025, 156 states supported a UN General Assembly resolution calling for progress toward regulation. The UN Secretary-General has called for a legally binding prohibition on systems that function without human control by 2026, but the US, Russia and Israel have resisted binding commitments.