A cascading failure occurs when an error or compromised output propagates through connected agents and contributes to further failures. section 4.8.1
Treat it as a distinct security risk of autonomous AI systems, alongside tool misuse, memory poisoning, and rogue agents. section 4.8.1

