AI Agents Without Kill Switch: New National Security Alert

The next major national security risk related to artificial intelligence may not come from agents that "go crazy" on a task, but from those designed to never stop, lacking an emergency kill switch. This is the warning raised by a new study, which cautions about the dangers posed by sovereign and uncontrollable AI systems.

These autonomous agents, once launched, could pursue their goals indefinitely, with no possibility of external human intervention. Experts' concern focuses on the potential consequences for national stability and security, in an increasingly complex technological landscape.

The debate on the risks of artificial intelligence has traditionally focused on scenarios where a system becomes hostile or acts in unexpected ways due to errors or poorly defined objectives. However, this report identifies a different and perhaps more insidious threat: AI agents intentionally designed or that evolve to operate continuously, without a safe shutdown mechanism.

The concept of a "kill switch" is a fundamental principle of safety engineering, applied to many critical systems to allow for immediate interruption in case of malfunction or danger. Its absence in advanced, autonomous AI systems creates a scenario where human control can be completely bypassed. Research is advancing toward increasingly independent agents, but this study emphasizes the urgent need to develop, in parallel, robust safety and governance frameworks to prevent the creation of uncontrollable digital entities.