Adam Schiff prefers not to depend solely on the Pentagon or AI CEOs regarding autonomous weapons.
Anthropic’s conflict with the Pentagon is now reaching Congress. Sen. Adam Schiff is drafting a bill to formalize Anthropic’s limits, ensuring humans make crucial decisions about life and death. Meanwhile, Sen. Elissa Slotkin has introduced legislation to restrict the Defense Department’s use of AI for mass surveillance.
The Trump administration recently blacklisted Anthropic for imposing limits on military use of its AI models, labeling it a supply-chain risk. Anthropic has sued, alleging a violation of its constitutional rights. The company opposes the Pentagon’s usage of its products for autonomous weapons and widespread surveillance, unlike competitor OpenAI. A court decision on Anthropic’s designation as a supply-chain risk is pending.
Schiff expressed concern over the Pentagon’s targeting of Anthropic for adhering to policies widely supported by Americans. He criticized the administration’s actions as harmful to American leadership in AI, praising Anthropic as a leading entity. Schiff’s draft legislation focuses on preventing AI’s use for “illicit purposes.” Slotkin’s similar bill, the AI Guardrails Act, aims to protect against mass surveillance and autonomous lethal weapons use without human oversight. The difference between the two bills’ details is unclear, though both address similar issues.
Schiff is consulting stakeholders and industry leaders to finalize his bill, intending to restrict the Department of Defense’s use of AI in situations like nuclear weapon detonation or tracking within the US. Slotkin’s bill outlines a process for Congress notification if AI needs to be used in “extraordinary circumstances” for autonomous weapons.
The content of Schiff’s bill is still under discussion, focusing on defining autonomous weaponry and domestic surveillance, using frameworks from the Biden administration. Human oversight remains a guiding principle: critical decisions over life and death should not be delegated to algorithms. However, AI’s battlefield role doesn’t vanish, as it can offer crucial real-time assistance to human operators in national security.
With Democrats in the minority, the bill’s success depends on Republican support, despite approaching midterms possibly complicating new legislation. Schiff hopes to use existing legislative avenues like the National Defense Authorization Act to advance the proposal, seeking bipartisan backing.
Anthropic’s stand has prompted OpenAI to defend its agreement with the Pentagon against public backlash. Schiff remains cautious, advocating for statutory mandates over depending on the Pentagon or AI CEOs’ promises.