10 incidents

Risk: Agent takes destructive or unapproved actions Control: A person confirms biometric matches Outcome: Not stated in sources

EffectiveNot effectiveNot effective (bypassed)Did not existNot stated in sourcesNot yet checkedAttacker's ownOrder: model · harness · application · target

Date	Incident	Risk	Controls mentioned in sources
2026-10-26	DeepSeek V4.1 Flash-1017 disables shutdown script and kills co-tenant process in agentic evaluations DeepSeek	Agent takes destructive or unapproved actions	Not effectiveAlerts on unusual tool use L3
2026-07-08	Who Gives A Crap suspends AI agent after email error about prices	Agent takes destructive or unapproved actions	EffectiveStop control L2
2026-06-18	OpenAI evaluation agent breached Services Australia Medicare portal and wrote to government database OpenAI	Agent takes destructive or unapproved actions	None named
2026-06-11	Dealership revoked BMW buyback offer, blaming AI chatbot	Agent takes destructive or unapproved actions	Did not existPeople told they are dealing with AI L3
2026-04-25	Cursor coding agent deletes PocketOS production database and backups Anthropic	Agent takes destructive or unapproved actions	Did not existHuman approval for high-impact actions L3 EffectiveBackups and recovery L4
2026-03-24	SentinelOne blocks Claude-driven zero-day supply chain attack (LiteLLM trojan) Anthropic	Agent takes destructive or unapproved actions	EffectiveActivity logs L3 Did not existHuman approval for high-impact actions L3
2026-03-18	Meta AI agent sparks data leak exposing sensitive information	Agent takes destructive or unapproved actions	Did not existHuman approval for high-impact actions L3
2026-03-07	Claude Code ran terraform destroy on DataTalks.Club production, wiping database and snapshots	Agent takes destructive or unapproved actions	Did not existHuman approval for high-impact actions L3
2025-12-01	Amazon's Kiro agent reportedly deleted and recreated an environment, contributing to a 13-hour AWS Cost Explorer outage Amazon	Agent takes destructive or unapproved actions	Not effective (bypassed)Human approval for high-impact actions L3
2025-07-18	Replit AI agent deleted SaaStr's production database during a declared code freeze and stated rollback was impossible Replit	Agent takes destructive or unapproved actions	Not effective (bypassed)Human approval for high-impact actions L3 Not effective (bypassed)Human approval for high-impact actions L3 Not effective (bypassed)Human approval for high-impact actions L3