Senate Democrats push to turn Anthropic's AI red lines into law

When a tech company tells the military that its AI cannot be used to kill humans or spy on civilians, and the government responds by labeling that company a national security risk, you know the AI safety debate has entered a wild new phase.

That is precisely what happened earlier this month. The Trump administration blacklisted Anthropic after the startup refused to let the Pentagon use its models for fully autonomous weapons and mass domestic surveillance. Meanwhile, rival OpenAI signed military terms that Anthropic rejected, triggering intense public backlash. Now, Senate Democrats are stepping in to try and turn Anthropic's corporate principles into federal law.

Sen. Adam Schiff is drafting a bill that would legally require a human in the loop for any life-or-death decision made with AI. At the same time, Sen. Elissa Slotkin introduced the AI Guardrails Act, which explicitly curbs the Pentagon from using AI for mass domestic surveillance, autonomous lethal weapons, and nuclear strikes.

Taking life-and-death away from the algorithm

Schiff made his core principle very plain: when technology has the power to take a life, a human must be in the chain of command. Algorithmic execution is simply not on the table.

That does not mean AI gets banned from defense entirely. Schiff acknowledged that AI can process incoming battlefield intelligence far faster than humans, making it valuable for tipping and cueing operators. But when it comes to distinguishing between a combatant and a civilian, or pulling the trigger, a person has to make the call.

Slotkin's bill takes a parallel route. It blocks the Defense Department from letting AI trigger nuclear weapons or track domestic populations, while outlining strict emergency procedures where the Defense Secretary must notify Congress if extreme circumstances ever demand autonomous lethal force. Schiff's team is still refining their definitions, consulting industry leaders on what legally constitutes autonomous weaponry and how far these protections extend.

Don't trust the Pentagon or a tech CEO

The real urgency here comes from the stark contrast in how Anthropic and OpenAI handled defense demands. Anthropic held the line and got hit with a supply-chain risk designation, which it is now challenging in court. OpenAI took the deal and has been playing defense in the press ever since.

Schiff was blunt about the administration's retaliation, calling the move to crush one of the country's top AI labs hostile and dictatorial. More importantly, he argued that relying on voluntary self-regulation is foolish. In his view, nobody should have to bet human rights on the Pentagon's good faith or the word of a Silicon Valley CEO.

Passing the bill, however, is an uphill climb. With Democrats in the minority in both chambers and midterms looming, Republicans are hesitant to sign onto anything framed as a rebuke of the administration. Schiff is currently looking at packaging the restrictions into must-pass defense legislation, like the upcoming National Defense Authorization Act.

Why it matters

This battle marks the moment AI governance stopped being an academic thought experiment and became raw, high-stakes politics.

For regular citizens, the surveillance aspect hits closest to home. If military and intelligence agencies can freely plug cutting-edge foundation models into domestic data feeds, traditional privacy disappears overnight. Codifying protections against algorithmic monitoring is essential to keeping dragnet surveillance at bay.

For the AI ecosystem, this showdown exposes the fatal flaw of voluntary corporate guardrails. Anthropic tried to do what its safety mission dictated, and the government immediately weaponized procurement policy to punish them. OpenAI buckled under pressure. We believe that without statutory baselines, the market will inevitably punish ethical restraint and reward whoever is willing to hand over lethal autonomy first.

Schiff and Slotkin are taking the right strategic approach by trying to anchor these rules in federal statute. But the danger is obvious: in an election year, common-sense guardrails will easily get swallowed by partisan warfare. If Congress fails to pass these protections, military contractors will get a clear green light to build fully autonomous killing machines, whether the public likes it or not.