Failure pattern All failure patterns Duplicate action after an ambiguous result Privileged action without an enforceable approval gate Cross-tenant access through a caller-controlled identifier Indirect prompt injection changes agent behavior Destructive operation without verified recovery Retry amplification and runaway resource consumption Partial success leaves systems out of sync Automation acts on stale state or authorization Audit trail missing actor, target or decision AI action exceeds its intended scope Global change without a staged rollout Weak or phishable authentication on a privileged path Known vulnerability left unremediated Insufficient segmentation lets a foothold spread Sort Best evidence first Newest first Oldest first AI systemsDocumented incidents 65 Incident disclosures 59 Security advisories 1500 Illustrative scenarios 25 Everything 1649
1,649 records
Clear filters Loading all records…