AI Guardrails Act: a senator wants to bar AI from deciding alone on lethal force or nuclear launch

On April 4, 2026, Michigan Democratic Senator Elissa Slotkin introduced the AI Guardrails Act of 2026. The bill bans three things: AI-aided mass surveillance, any autonomous lethal action taken without human validation, and any nuclear-launch decision made independently by an AI system. The senator plans to push these guardrails through the next National Defense Authorization Act rather than as a standalone bill.

The introduction follows a public dispute between Anthropic and the Pentagon, which had designated the company a supply-chain risk before cutting its contracts. Slotkin's bill does not target a specific vendor: it sets a structural limit on AI autonomy in decisions with irreversible consequences, whether lethal targeting or nuclear deterrence.

Cardan-AI analysis: the strongest case for the bill is economic, not ethical. An irreversible decision — killing, launching a nuclear weapon — destroys the possibility of waiting for better information before acting. Irreversibility theory (Arrow & Fisher, 1974) shows that this lost option has a quantifiable cost, distinct from the cost of the error itself: it is the 'option value' sacrificed when a machine is allowed to decide faster than a human could verify the information.

For aerospace and defense contractors under federal contract, the signal is clear even before any eventual passage: documenting where, in the lethal decision chain, an irreducible human validation point is architected — and being able to demonstrate it under audit — is becoming a compliance prerequisite, not a design option.