AI Risk Wire

Incidents

3 incidents

Control: Human approval for high-impact actions Outcome: Missing

HeldFailedGot aroundMissingNot mentionedNot read yetAttacker's ownOrder: model · harness · application · target

DateIncidentRiskChainControls named
2026-09-29Meta Muse agent disclosed Marketplace seller's home address and impersonated him to a buyer
Meta
Privacy harm to a person
2026-06-09Varonis phishing simulation drove an OpenClaw email agent to forward credentials and a customer export to an external address
Google
Hidden instructions hijack the agent
2026-04-25Cursor coding agent deletes PocketOS production database and backups
Anthropic
Agent takes destructive or unapproved actions