Dear Readers,

The U.S. government threatened to blacklist one of the world's leading AI companies - not for espionage, not for sabotage, but for refusing to remove its own safety guardrails. In today's feature, we unpack the full story behind the Anthropic–Pentagon standoff that has erupted since late February 2026: the ultimatum, the legal brinkmanship, and the moment OpenAI stepped in with a deal that looked like the same protections on the surface but may have been built on a very different foundation. We walk you through the contract clauses, the surveillance loopholes in existing law, and the uncomfortable question at the center of it all — who actually gets to draw the red lines for military AI? Then, in Chubby's Opinion Corner, we dig into the leaked Dario Amodei memo, the "safety theater" accusation, and what this fight could mean for Anthropic's IPO and the competitive balance of the entire AI industry. This one cuts deep — let's get into it.

All the best,

Disclaimer: Because we know that the topic is currently being hotly and emotionally debated, it is important for me as the author to say that the article is not about representing one side or the other, or about taking any particular position, but about analytically processing and presenting the available information about this situation in a way that is informative and useful to our readers.

“Everyone having a superintelligent genius in their pocket is an amazing advance and will lead to an incredible creation of economic value and improvement in the quality of human life. I talk about these benefits in great detail in Machines of Loving Grace. But not every effect of making everyone superhumanly capable will be positive. It can potentially amplify the ability of individuals or small groups to cause destruction on a much larger scale than was possible before, by making use of sophisticated and dangerous tools (such as weapons of mass destruction) that were previously only available to a select few with a high level of skill, specialized training, and focus.”

— Dario Amodei, The Adolescence Of Technology

Imagine a world where the most powerful artificial intelligence systems on the planet are deployed by the military with no hard limits on what they can and cannot do. No explicit prohibition against turning AI loose on the personal data of American citizens. No categorical ban on letting algorithms make life-or-death targeting decisions without a human in the loop. That scenario is not hypothetical. It is, in essence, what the U.S. Department of War demanded from one of America’s leading AI companies, and what triggered the most consequential clash between Silicon Valley and the Pentagon in the short history of frontier AI.

In late February 2026, the conflict between Anthropic, the maker of the Claude AI system, and the newly rebranded Department of War (DoW) — formerly Department of Defense (DoD) —erupted into full public view. Anthropic had drawn two bright lines in its government contracts: its technology would not be used for mass domestic surveillance of U.S. persons, and it would not power fully autonomous weapons - systems that remove humans entirely from targeting and engagement decisions. The DoW, operating under a January 2026 strategy memo signed by Secretary of War Pete Hegseth, wanted something fundamentally different. It wanted AI contracts built on “standard ’any lawful use’ language,” with no vendor-imposed constraints that might slow down military adoption. When Anthropic refused to budge, the government didn’t just walk away. It threatened to designate Anthropic a “supply chain risk” - a label normally reserved for adversarial foreign actors trying to sabotage national security systems - and moved to ban the company’s products across federal agencies.

Within days, OpenAI stepped in, signing a classified-network agreement with the DoW and publishing its own set of “red lines.” But here is the question that has since consumed the AI policy world: Did OpenAI actually secure the same protections Anthropic was willing to go to court over, or did it simply find a more palatable way to say yes?