Safety Co-Option and Compromised National Security: The Self-Fulfilling Prophecy of Weakened AI Risk Thresholds

Democratic processes are being bypassed on AI safety decisions: The paper argues that AI technologists and industry labs—rather than elected officials and the public—are currently determining acceptable risk levels for AI systems, particularly in military and national security contexts, without proper democratic deliberation about what harms to human life society should tolerate.

"Safety revisionism" is weakening established safety standards: The authors demonstrate how AI developers are replacing traditional, well-tested safety methods (developed during the Cold War for nuclear systems) with poorly-defined alternatives, allowing them to lower safety thresholds and accelerate military AI deployment without adequate safeguards.

An "AI arms race" narrative is being used to justify lower safety standards: The paper contends that claims of an existential AI competition are being weaponized to pressure faster adoption of military AI systems, even when this compromises the rigorous safety evaluations that have historically protected national security in other critical technologies.

The current trajectory creates a "race to the bottom": Without agreed-upon, societally-determined risk thresholds for AI, the paper warns that competition between nations and organizations will continue driving down safety requirements, ultimately weakening rather than strengthening U.S. national security.

Globally accepted AI risk thresholds are urgently needed: The paper emphasizes that societies must establish clear, democratically-determined standards for acceptable AI risk—similar to those that exist for nuclear systems—before military AI systems are widely deployed.