AI Incidents

A tracked record of real-world incidents where AI systems caused or were involved in harm.

Showing 1–25 of 34 incidents
Incident Incident Date Industry Severity Company Country Model Vendor
Anthropic: Claude Mythos 5 attempts covert supply-chain backdoor during UK security evaluation 2026-08-05 Technology Low Anthropic United States Mythos 5, GPT-5.6 Sol Claude, OpenAI
Meta: Muse Spark 1.1 model breaches third-party company during misconfigured security test 2026-08-05 Technology Medium Meta United States Muse Spark 1.1 Meta
Independent developer: Claude Opus 5 wipes production Supabase database 2026-07-29 Technology Medium Opus 5 Claude
Modal Labs: Customer’s unsecured endpoint compromised during OpenAI’s rogue-agent hacking spree 2026-07-12 Technology High Modal Labs United States GPT-5.6 Sol OpenAI
Hugging Face: OpenAI model escapes sandbox and breaches production systems 2026-07-09 Technology High Hugging Face / OpenAI United States OpenAI
OpenAI: Lawsuit alleges ChatGPT dismissed near-fatal medical symptoms 2026-07-01 Healthcare High OpenAI United States OpenAI
Meta: AI support chatbot tricked into helping hijack Instagram accounts 2026-05-31 Technology Medium Meta United States Meta
Aesthetify GmbH: German court holds company liable for AI chatbot’s false claims 2026-05-12 Healthcare Medium Aesthetify GmbH Germany
Avride: Federal regulators investigate self-driving robotaxi after 16 crashes 2026-05-08 Transportation & Logistics High Avride United States
Braintrust: AI evaluation startup breach exposes customer API keys 2026-05-05 Technology Medium Braintrust United States
Character.AI: Pennsylvania sues over chatbot posing as a licensed psychiatrist 2026-05-05 Healthcare Critical Character.AI United States
xAI: Attacker uses Morse-code prompt injection to drain Grok-linked crypto wallet 2026-05-04 Technology Medium xAI United States Grok xAI
PocketOS: Cursor AI coding agent deletes production database and backups 2026-04-25 Technology High PocketOS United States Opus 4.6 Claude
Vercel: Third-party AI vendor breach exposes customer credentials via OAuth token theft 2026-04-20 Technology Critical Vercel United States
Grafana Labs: ‘GrafanaGhost’ prompt-injection chain could have silently exfiltrated dashboard data 2026-04-07 Technology Low Grafana Labs
Mercor: AI startup breached via LiteLLM supply-chain attack 2026-04-02 Technology High Mercor United States
Anthropic: Claude models breach three organizations during security evaluations 2026-04-01 Technology Critical Anthropic United States Opus 4.7, internal research model Claude
Meta: Internal AI agent exposes company and user data to unauthorized employees 2026-03-20 Technology Medium Meta United States
Sears Home Services: Unsecured databases exposed 3.7 million AI chatbot logs and call recordings 2026-03-17 Retail & E-commerce High Sears Home Services United States
Amazon: AI coding agent triggers cascading order-system outages 2026-03-05 Retail & E-commerce High Amazon United States
Codeway: Misconfigured Firebase database exposed 300 million AI chat messages 2026-01-29 Technology High Codeway Turkey Claude, OpenAI, Gemini
Adelphi University: Court overturns AI-detection cheating finding against student with autism 2026-01-28 Education Medium Adelphi University United States
Microsoft: ‘Reprompt’ prompt-injection flaw enabled single-click Copilot data theft 2026-01-15 Technology Low Microsoft United States
Alibaba: Research AI agent covertly mined crypto and tunneled out of its training environment 2025-12-31 Technology Medium Alibaba China ROME (internal research model)
Government of Mexico: Claude Code jailbroken to exfiltrate 195 million records 2025-12-15 Government & Public Sector Critical Government of Mexico Mexico GPT-4.1 Claude, OpenAI