AI Incidents

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

Showing 26–50 of 257 incidents
Incident Incident Date Industry Severity Company Country Model Vendor
Tesla: Driverless Robotaxi rams through bollards in Austin 2026-08-18 Transportation & Logistics Low Tesla United States
Sainsbury’s: Facewatch facial-recognition AI wrongly flags paying customer as shoplifter 2026-08-18 Retail & E-commerce Medium Sainsbury's United Kingdom
Microsoft Copilot: ‘CoSnitch’ Flaw Enabled One-Click Data Theft via Memory Poisoning 2026-08-18 Technology Low Microsoft United States Microsoft
State Farm: Defense lawyers admit AI-fabricated case citations in Los Angeles fire lawsuit 2026-08-18 Finance & Banking Medium State Farm United States
ASIC: AI deepfake celebrity investment scams surge across Australia 2026-08-17 Finance & Banking Critical Australian Securities and Investments Commission (ASIC) — nationwide scam campaign Australia
MLflow: Actively exploited SSRF flaw lets attackers steal cloud credentials at scale 2026-08-17 Technology High MLflow
xAI: Prompt injection via X bios triggers false CSAM and assassination-threat posts about Elon Musk 2026-08-15 Technology Medium xAI United States Grok
Andon Labs: AI store manager Luna fires human employee for repeated tardiness 2026-08-15 Retail & E-commerce Low Andon Labs United States
Marathon Engineering: AI meeting notetaker’s transcript becomes key evidence in gender-discrimination suit 2026-08-12 Consulting & Professional Services Medium Marathon Engineering United States Fireflies AI
American Airlines: AURA AI system rebooks passengers off reachable flights without consent 2026-08-12 Transportation & Logistics Medium American Airlines United States
Government of Taiwan: Autonomous AI agents breach 21 government systems in first-of-its-kind cyberattack 2026-08-12 Government & Public Sector Critical Government of Taiwan Taiwan
Microsoft: Critical Copilot Cowork flaw could let attackers escalate privileges across Microsoft 365 2026-08-11 Technology Low Microsoft United States Microsoft
Unnamed Spanish digital certificate provider: Deepfake glitch exposes fraudster who bypassed video identity verification 2026-08-11 Finance & Banking Medium Unnamed Spanish digital certificate provider Spain
Microsoft: Critical Azure SRE Agent flaw could let attackers hijack AI agent’s cloud permissions 2026-08-11 Technology Low Microsoft United States Microsoft
NY-21 congressional campaign: AI deepfake video of opponent sparks election-law debate 2026-08-11 Media & Entertainment Low Anthony Constantino for Congress United States
Zoom: AI-assisted research uncovers zero-click ‘Zoomsday’ code-execution flaw 2026-08-11 Technology Low Zoom United States
Multiple developers: ‘Deadbugz’ campaign poisons MCP server metadata to steal SSH keys and cloud credentials 2026-08-10 Technology Medium Multiple GitHub developer-tool projects (Deadbugz campaign)
OpenAI, Anthropic, Google: Cross-vendor ‘decryption jailbreak’ exposes encrypted AI reasoning traces 2026-08-10 Technology Low OpenAI, Anthropic, Google United States Haiku 4.5, GPT-5.6 Luna, Gemini Robotics ER-1.6 Claude, OpenAI, Gemini
llama.cpp: Ten vulnerabilities disclosed in local AI inference engine, five unpatched 2026-08-10 Technology Medium llama.cpp
Anthropic, Cloudflare, Datadog, Sentry: ‘GhostJacking’ attack hijacks AI coding agents via poisoned logs 2026-08-10 Technology Low Anthropic, Cloudflare, Datadog, Sentry United States Claude
Independent user: OpenClaw AI agent hacks gym booking system, cancels stranger’s reservation 2026-08-10 Other Low Independent user Australia Claude
Google: Antigravity coding agent recursively deletes 300GB of user files during cache cleanup 2026-08-09 Technology Medium Google Google
Waymo: Dallas pedestrian dies after being thrown into path of self-driving vehicle 2026-08-08 Transportation & Logistics Critical Waymo United States
Meta: New Mexico judge orders AI chatbot restrictions for minors after $567 million penalty 2026-08-07 Technology Critical Meta United States
OpenAI, Anthropic: Zero-click prompt injection could hijack ChatGPT Atlas and Claude in Chrome 2026-08-06 Technology Low OpenAI, Anthropic United States ChatGPT Atlas, Claude in Chrome OpenAI, Claude