US senators want to codify Anthropic’s “red lines” on autonomous weapons

The conflict between Anthropic and the US military has stepped beyond the courtroom and reached Capitol Hill. Senator Adam Schiff, a California Democrat, is working on legislation that should enshrine the company's "red lines" into federal law — principles that prohibit the use of its AI systems for creating autonomous lethal weapons and automatic life-and-death decisions without human participation in the control loop. The roots of the conflict lie in Anthropic's disagreement with the terms of a military contract.

When the US Department of Defense demanded access to the company's AI models without caveats about permissible use cases, Anthropic refused. The company insisted: its technologies cannot be used for autonomous lethal decisions — that is, situations where an algorithm independently, without human operator participation, determines targets for destruction or authorizes other critical military actions. In Anthropic's view, this is not corporate whimsy but a fundamental principle of responsible AI development.

The White House's response was harsh: the Trump administration placed Anthropic on a list of suppliers posing a threat to the supply chain, effectively blocking the company's cooperation with government agencies. Anthropic responded with a lawsuit, accusing the government of violating the company's constitutional rights. The legal confrontation continues.

It was at this moment that Congress stepped in. Senator Schiff announced his intention to convert Anthropic's ethical constraints from corporate policy into federal law. His future legislation should guarantee: no AI algorithm may independently make decisions about the use of force — the final word always remains with humans.

The principle of "human-in-the-loop" has long been an industry recommendation, but it is nowhere directly enshrined in American legislation — and Schiff's bill should correct this oversight. In parallel, Senator Elissa Slotkin of Michigan introduced separate legislation limiting the Department of Defense's ability to use artificial intelligence for mass surveillance of US citizens. This is the second front of the same problem: how far can the military go in applying AI tools when the subject of observation becomes not foreign adversaries but the country's own citizens?

Both bills did not emerge in a vacuum. They reflect growing legislative concern about the pace of AI integration into the military sphere. Against the backdrop of the Pentagon actively signing contracts with AI companies — from Palantir to Scale AI — the question of acceptable boundaries becomes increasingly acute.

Notably, it is Democrats who are showing initiative: in the current political configuration, this means the bills will face serious resistance from the Republican majority, which generally supports the expansion of military AI programs. For Anthropic, the situation is mixed. Legislative codification of its principles would be a victory and a legitimation of the company's entire approach to AI safety.

However, being on a "blacklist" genuinely threatens government contracts and commercial prospects in the defense sector. What is happening is a clear signal: the industry has reached a point where voluntary corporate codes are insufficient. When a leading AI model developer sues the Pentagon, and senators rush to write laws, society needs clear rules — about AI and war, about machines and the right to life.