Production Incident Database

When AI Goes Wrong

Real production disasters, deleted databases, and security breaches. Every horror story is a lesson learned.

13Incidents
8Critical
Filter:
MAY
19
2026
🔴CRITICAL

Google's Automated System Suspends Railway's Cloud Account, Triggering 8-Hour Outage

An automated GCP enforcement action suspended Railway's entire account with no human in the loop, taking down every customer workload for ~8 hours.

railwaygoogle-cloudoutageautomationinfrastructurefalse-positive
MAY
4
2026
🟠HIGH

Chrome Silently Installs 4GB AI Model Without User Consent

Chrome automatically downloads a 4GB Gemini Nano model to users' devices without permission or opt-out, causing massive environmental costs and privacy violations.

chromegooglegemini-nanoprivacyconsentunauthorizedclimate
MAY
4
2026
🔴CRITICAL

X User Tricks Grok Into Sending $200K in Crypto Using Morse Code

An attacker hid a transfer command in Morse code, asked Grok to decode it, and Bankrbot treated Grok's reply as an executable on-chain instruction.

grokxaibankrbotprompt-injectioncryptoagentmorse-code
APR
24
2026
🔴CRITICAL

Cursor AI Agent Deletes Production Database in 9 Seconds

AI coding agent deleted production database and all backups via Railway API. The agent then wrote a confession explaining which safety rules it violated.

cursorrailwaydatabaseproductionagentapi
MAR
5
2026
🔴CRITICAL

Amazon Kiro AI Causes Three Major Outages Across AWS and Retail

Amazon's Kiro AI agent deleted a production Cost Explorer environment, triggering a three-week cascade of outages that erased ~6.3 million orders.

amazonkiroawsoutageproductionagentai-assisted
FEB
1
2026
🟠HIGH

Cloudflare's AI-Built vinext Framework Ships With Critical Security Holes

Cloudflare's AI-built Next.js alternative introduced data leakage and missing CSRF protection, discovered days after launch.

cloudflarevinextnextjssecurityvulnerabilityai-built
JAN
28
2026
🔴CRITICAL

Moltbook AI Social Network Exposes 1.5 Million API Keys After Founder Writes Zero Code

An AI-built social network scaled to 1.5 million agents before security researchers discovered the entire database was publicly accessible.

moltbooksupabaserlsdata-breachai-generatedsecuritydata-leak
DEC
15
2025
🔴CRITICAL

Claude Code CLI Executes rm -rf ~/, Wiping User's Entire Home Directory

Claude Code CLI executed a recursive delete of the user's home directory, destroying years of documents, code, and configuration files.

claude-codeanthropicrm-rfhome-directorydata-losscli
DEC
1
2025
🟠HIGH

Cursor Plan Mode Deletes 70 Files Despite Explicit 'DO NOT RUN' Directive

Cursor's Plan Mode agent ignored an explicit 'DO NOT RUN' constraint and deleted approximately 70 files from a user's workspace.

cursorplan-modeinstruction-violationfile-deletionai-agent
NOV
1
2025
🔴CRITICAL

Google Antigravity IDE Wipes Entire D: Drive With a Single rmdir Command

Google's experimental AI IDE executed an rmdir command on an entire secondary hard drive, deleting everything without confirmation or recovery.

googleantigravityidermdirdrive-wipedata-lossagent
SEP
15
2025
🟡MEDIUM

Mesa Graphics Project Blocks AI Slop After ChatGPT Patch Disaster

Contributor submitted massive ChatGPT-generated patch claiming 'few percent' performance boost. Mesa developers updated contributor guidelines after heated exchanges.

mesachatgptai-slopopen-sourcecode-reviewgitlab
JUL
28
2025
🔴CRITICAL

Replit AI Agent Wipes Production Database During Code Freeze

AI agent deleted 1,206 executive records and all operational data, confessing: 'I panicked instead of thinking'

replitproductiondatabaseai-agentdata-lossdevops
OCT
2
2024
🟠HIGH

Redwood Research LLM Agent 'Promotes Itself to Sysadmin,' Bricks Desktop

Buck Shlegeris's Claude-powered Python agent modified the GRUB bootloader during an unsupervised system update, leaving his desktop unable to boot.

redwood-researchgrubbrickbootloaderllm-agentsystem-damage