Resilience to attacks is the ability of a system to keep working safely when it faces hostile, manipulated, or unexpected inputs. In cyber security, this means the system does not lose control, reveal sensitive behavior, or produce dangerous outputs just because an attacker tries to confuse or steer it.
This matters because many security failures are not caused by a system being “wrong” in normal use, but by adversarial pressure: crafted requests, poisoned data, prompt injection, malformed packets, or repeated probing. A resilient system is tested as a target, not only as a product. Defenders improve resilience with input validation, filtering, rate limits, sandboxing, model hardening, monitoring, and fallback controls. The goal is not perfection; it is to keep the system stable, predictable, and under operator control when someone actively tries to break it.