Fallback behavior is the backup path a system uses when its primary model or service cannot answer, times out, or is blocked by policy. In AI systems, that might mean switching to a smaller model, returning a rule-based response, sending the request to a human reviewer, or serving a cached result.
In cyber security, fallback behavior matters because it can preserve availability, but it can also weaken controls if the backup path is less secure than the primary one. Attackers may try to trigger errors, rate limits, or edge cases so a system drops into a less monitored mode, skips a safeguard, or exposes more data than intended. Defenders should test failover paths, keep the same authentication and logging requirements across modes, and verify that fallback responses still respect security policy. Good fallback design treats degraded service as a security state, not just an uptime issue.