Human fallback is a control pattern that sends uncertain, risky, or high-impact actions to a person before the system proceeds. In AI-driven workflows, it is the point where automation stops and review begins. The goal is not to block automation, but to keep judgment where the consequences are hardest to reverse.
In cybersecurity, human fallback matters because agents and automated tools can misread context, overreach permissions, or take unsafe actions at machine speed. A secure design uses fallback for events such as privilege changes, repository writes, deployment approvals, payment steps, or responses to ambiguous alerts. Defenders also use it to verify suspicious outputs, confirm detections, and prevent accidental or malicious changes. Without a human fallback, an attacker who manipulates an AI workflow may be able to turn a small prompt or false signal into an unauthorized action.