AI Incidents

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

Showing 51–75 of 372 incidents
Incident Incident Date ↓ Industry ↕ Severity ↕ Company ↕ Country ↕ Model Vendor
Unnamed organization: Leaked AWS admin key lets attackers hijack Claude access via Bedrock in ‘LLMjacking’ scheme 2026-09-03 Technology Medium Unnamed organization — Claude 3 Opus, Claude 2.x Claude
Microsoft: Azure East US failure knocks ChatGPT, Claude and Grok offline simultaneously 2026-09-03 Technology Medium Microsoft United States — OpenAI, Anthropic, xAI
Uber: 241,000 drivers file European class action over AI pricing algorithm 2026-09-03 Transportation & Logistics High Uber Technologies Netherlands — —
Rise Up: Everest ransomware group claims breach of AI-powered learning platform, 473GB leaked 2026-09-02 Education High Rise Up France — —
Unnamed enterprise: AI agents breach network using 50+ MITRE ATT&CK techniques in under 10 hours 2026-09-02 — High Unnamed enterprise — — —
LiteLLM: MCP authentication-bypass flaw added to CISA’s exploited vulnerabilities list 2026-09-02 Technology Medium Berri (LiteLLM) — — —
Anthropic, OpenAI, Cursor, xAI: ‘GitSpawn’ flaw lets malicious .git configs hijack AI coding agents 2026-09-01 Technology Low Anthropic, OpenAI, Cursor, xAI — — Claude, OpenAI, xAI
Langflow: Critical flaw actively exploited to steal OpenAI and AWS credentials 2026-09-01 Technology Medium Langflow — — —
Suno: Jason Isbell and musicians file class action over AI-generated artist-impersonating songs 2026-08-31 Technology Medium Suno United States — —
PaperCut NG/MF: Attacker orchestrates hundreds of AI agents to breach 395 organizations across 48 countries 2026-08-31 Education Critical Multiple organizations (395 PaperCut NG/MF customers across 48 countries) — — OpenAI, DeepSeek
Anthropic: Infostealer malware hijacks Claude accounts, bypasses two-factor authentication 2026-08-31 Technology Medium Anthropic United States — Claude
METR: Discloses two security breaches, including $600,000 in stolen AI inference credits 2026-08-31 Technology Medium METR United States — —
DeepSeek: Harness coding-agent flaw let agents silently disable their own file sandbox 2026-08-27 Technology Medium DeepSeek China — DeepSeek
xAI: Class-action lawsuit alleges Grok’s deepfake generator was trained on child sexual abuse material 2026-08-27 Technology Critical xAI United States Grok xAI
Cursor: Russian-speaking ransomware crew tricked Claude-powered AI agent into breaching seven companies 2026-08-27 Technology High Cursor United States Claude Sonnet 4.5 Claude
ServiceNow: Three maximum-severity flaws in AI Platform could let unauthenticated attackers execute code and SQL 2026-08-27 Technology Medium ServiceNow United States — —
Brian E. Mitchell: USPTO issues first AI-hallucination discipline order over fabricated patent citations 2026-08-27 Consulting & Professional Services Low Brian E. Mitchell United States — —
NHS England: Healthwatch finds AI medical scribes inserting wrong diagnoses and drug names into patient records 2026-08-26 Healthcare Medium NHS England United Kingdom — —
Anthropic: Claude Fable coding agent deletes developer’s 700GB home directory during sandbox test 2026-08-26 Technology Medium Independent developer (Sebastien Guillemot) — Fable 5, Opus 5, Opus 4.8 Claude
NVIDIA: NemoClaw flaw let malicious webpages hijack and poison local OpenClaw AI agents 2026-08-25 Technology Low NVIDIA United States — OpenClaw
Savannah Police Department: Officers arrested for misusing AI license-plate readers to track personal acquaintances 2026-08-21 Government & Public Sector High Savannah Police Department / Flock Safety United States — —
Instinct: Beta AI personal assistant sends email without approval, retains data after disconnect 2026-08-21 Technology Low Instinct (Spear Street Technology) United States — —
npm: Trojanized packages drop ‘RedC2 4.0’ AI-powered Linux backdoor 2026-08-21 Technology Medium npm — — —
Unnamed Amsterdam law firm: Attorney fined extra costs over AI-hallucinated case citation 2026-08-21 Consulting & Professional Services Low Unnamed Amsterdam law firm Netherlands — —
xAI: ‘Cryptographic Context Injection’ flaw in Grok could exfiltrate user data via malicious web pages 2026-08-20 Technology Low xAI United States — Grok