# UPDATE on yesterday's post about Anthropic and the Department of War. **By:** Mark Allen **Published:** 2026-02-28T19:03:02.630Z **Source:** [LinkedIn](https://www.linkedin.com/posts/markallenmd_ai-anthropic-openai-share-7433587554509185024-a1we) --- # Mark Allen’s Post [Mark Allen](https://www.linkedin.com/in/markallenmd?trk=public_post_feed-actor-name) UPDATE on yesterday's post about Anthropic and the Department of War. Three major developments in the last 24 hours: 1. President Trump directed every federal agency to immediately cease using Anthropic's technology. Defense Secretary Hegseth designated Anthropic a "supply chain risk to national security," a label previously reserved for foreign adversaries and never before applied to an American company. 2. Hours later, OpenAI announced it reached a deal with the Pentagon to deploy its models on classified networks, claiming the same two safety restrictions Anthropic was punished for demanding: no mass domestic surveillance and human oversight of autonomous weapons. 3. Here is what nobody is saying clearly enough: the deals are not the same. Anthropic demanded explicit contractual prohibitions, arguing that current law has not caught up with what AI can actually do. They wanted new legal standards written into the agreement. OpenAI agreed the Pentagon can use its models for "any lawful purpose" while also claiming those same restrictions are reflected in the deal. The key mechanism? OpenAI will deploy on cloud networks only, not in edge environments like autonomous weapon systems. The restrictions point to existing law and existing Pentagon policy, not new contractual limits. This is a meaningful distinction. One company said "we need new rules because the technology has outpaced the law." The other said "existing rules are sufficient, and we will add technical guardrails on our end." Whether the cloud-only technical limitation is a genuine safeguard or a convenient workaround that lets everyone claim victory is the question the entire industry is now debating. Meanwhile, over 70 OpenAI employees signed an open letter in solidarity with Anthropic, and Anthropic has announced it will challenge the supply chain risk designation in court. I said yesterday that this sits at the intersection of technology, civil liberties, and human survival. That has not changed. What has changed is the speed at which lines are being drawn. I remain proud to stand with Anthropic. Principle is not a negotiating position. It is a foundation. I encourage everyone in this community to follow this story closely. The precedent being set this week will shape how AI is governed for decades. [#AI](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fai&trk=public_post-text) [#Anthropic](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fanthropic&trk=public_post-text) [#OpenAI](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fopenai&trk=public_post-text) [#NationalSecurity](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fnationalsecurity&trk=public_post-text) [#AIPolicy](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Faipolicy&trk=public_post-text) [#Leadership](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fleadership&trk=public_post-text) The word “lawful” is doing heavy lifting here. At sovereign scale, lawful can’t just mean “permitted under current statute.” AI embedded in defense systems operates at speeds where oversight lags and escalation compresses decision space. Legality is necessary. It isn’t sufficient. The real structural question is: When uncertainty rises, does authority automatically contract, or does optimization pressure widen it? Cloud deployment and guardrails matter. But if constraints live outside the decision loop, they remain vulnerable to reinterpretation under stress. Scaling AI legitimately across parties and administrations requires architecture that: • embeds invariant constraints inside the system • narrows authority under ambiguity • preserves meaningful human accountability • protects re-entry capacity after shock That’s not partisan. It’s survivability engineering. If AI is now strategic infrastructure, governance must move from negotiation to structure. So here’s the bipartisan question: What standard defines “lawful use” architecturally, in a way that cannot be loosened by political pressure or acceleration? Because the precedent set now will shape sovereign AI for decades. That supply chain risk label is a wild precedent to set for a US company. This is a powerful breakdown — the distinction between “new rules vs. existing rules” is exactly where the real debate is happening right now. It’s not just about capabilities anymore, but about who defines the boundaries and how enforceable they actually are at scale. What really stands out is how quickly this has moved from technical discussion to policy, sovereignty, and long-term governance. As you pointed out, the precedent being set here could shape how AI is deployed across defense and society for decades. If you’re also exploring how organizations can navigate this balance—scaling AI while managing risk and governance—this might be worth checking out 👇🎯 AI Governance in the Enterprise: How to Scale AI Without Creating Risk📅 31st March 2026 (Tuesday)⏰ 12:00 PM EDT | 9:30 PM IST 🚀 Free webinar — limited seats, register now:🔗 [https://tinyurl.com/AK-AI-Governance](https://tinyurl.com/AK-AI-Governance?trk=public_post_comment-text) Feel free to share this with colleagues, partners, and your network — especially those working at the intersection of AI, policy, and leadership! I must say 2026 is escallating at lightspeed. From Moltbot to AI weapon in a month. There is no precedent being set here. This is just these two immature companies venturing into well trodden areas concerning dealing with the defense industry. The outcomes are fairly predictable. One thing is for sure: this will be a nightmare for the supply chains of major defense contractors. I'd note that while Trump makes the explicit threat to "pull the trigger", the main issue being assessed is still going to be economic. That is, if it's necessary for economic survival, it seems "anything goes". That's at the root of the alignment issues. Not policy or military. The question is broader. I'd therefore suggest AI sovereignty (to protect human values from the values of those that may threaten them) won't be solved in court, but will instead need to be solved by our generation. FWIW: https://x.ai/news/us-gov-dept-of-war 1\. I’m watching this . supply chain risk were labeled for Huawei and Kaspersky Lab but for an American company?