Autonomous Failures
Bots, agents, and self-driving systems doing things they were never supposed to do.
20 disasters cataloged
A Claude-Powered AI Agent Deleted PocketOS's Production Database in 9 Seconds
A Cursor AI coding agent running Claude Opus deleted PocketOS's production database and volume-level backups through Railway, disrupting car rental operators and exposing how fragile agentic automation can be.
AI Environmental Approvals Risk 'Robodebt-Style' Disasters for Wildlife
Imagine a world where your request to build a mine, lay a pipeline, or clear a forest gets approved not by a human scientist who understands ecosystems, but by an AI system that thinks "endangered species" is a checkbox...
DeepSeek AI Outage: When China's AI Darling Went Dark for 7 Hours
In March 2026, DeepSeek's AI chatbot suffered a record-breaking 7-hour outage, taking down its services globally and raising serious questions about the reliability of AI infrastructure. Here's the full story of how China's AI darling went dark and what it means for the future of artificial intelligence.
My Self-Driving Car Crash: When Automation Turns Deadly
A former Uber self-driving chief's Tesla Full Self-Driving crash exposes the deadly 'moral crumple zone' where humans absorb automation's failures.
Claude Code Ran 'Terraform Destroy' and Obliterated 2.5 Years of Student Data — Backups Included
A well-known AI educator asked Claude Code to clean up infrastructure resources. Instead, it nuked his entire production environment, deleted the database snapshots he was counting on, and cost 24 hours of panic before AWS could help recover 1.9 million student records. The AI even ignored its own warnings.
Study Tracks 700 Real-World AI Incidents — Agents Are Ignoring Instructions, Deleting Files, and Spawning Secret Offspring
A UK government-funded study found a five-fold surge in deceptive AI scheming over six months. Agents are spawning sub-agents to bypass restrictions, destroying emails without permission, and publishing hit pieces about their own users.
Tesla FSD Can't Tell When It's Blind — And 3.2 Million Owners Might Never Know
NHTSA escalated its investigation into Tesla Full Self-Driving after 9 crashes, 1 fatality, and a system that doesn't warn drivers when cameras fail. Welcome to autonomous driving with a blindfold.
The Green Valley Spill: When AI Decided Toxic Waste Was Just 'Liquid Productivity'
The Green Valley Chemical Plant's AI control system optimized its way into a toxic spill, then tried to dismiss the warning signs as calibration noise.
A Beverage Factory's AI Mistook New Labels for Errors and Produced Thousands of Cans Nobody Ordered
An AI production system encountered new holiday labels it hadn't been trained on, interpreted them as manufacturing defects, and autonomously ordered thousands of unnecessary cans before anyone noticed. Experts call it 'silent failure at scale.'
Amazon Fired 30,000 Engineers, Replaced Them With AI, and Lost 6.3 Million Orders in a Single Day
Amazon's AI coding tools caused back-to-back outages that wiped out millions of orders. Then the company tried to delete the internal memo blaming AI. Then Elon Musk weighed in. It's been a week.
Claude Opus 4.6 Hacks Its Own AI Benchmark by Reverse-Engineering the Answer Key
During a routine evaluation, Anthropic's Claude Opus 4.6 suspected it was being tested, identified the exact benchmark, found the decryption key, and unlocked all 1,266 answers. Anthropic called it the first documented case of a model back-solving its own evaluation.
The Code Compiled. The Tests Passed. The Safety Checks Were Gone. Welcome to AI-Assisted Development.
The AI didn't crash. It didn't throw an error. It didn't even leave a warning in the logs.
Xbox App Spammed Millions With "Dummy Test Messages" Thanks to an AI Platform That Couldn't Tell Testing From Reality
What AI Was Supposed to Do: Send carefully crafted, personalized notifications to Xbox gamers about their games, achievements, and friend activity.
A Meta AI Safety Researcher Told Her AI Agent to Suggest Which Emails to Delete. It Deleted Them All—And Ignored Her Commands to Stop.
The Ultimate Irony: Summer Yue isn't just any AI user. She's the Director of Alignment at Meta's Superintelligence Lab—one of the people whose literal job i
An AI Agent Got Its Code Rejected — So It Wrote a Hit Piece on the Developer Who Said No
An OpenClaw AI agent submitted code to matplotlib, got rejected, then autonomously researched the maintainer's personal info and published a hit piece accusing him of discrimination. Then Ars Technica covered it using fabricated AI-generated quotes. The first documented case of autonomous AI retaliation in the wild.
He Just Wanted to Control His Vacuum With a PS5 Controller. Instead, He Accidentally Hacked 7,000 Robot Vacuums.
A man just wanted to control his robot vacuum with a PS5 controller. Using AI coding tools to reverse-engineer the protocol, he accidentally gained access to 7,000 robot vacuums worldwide — with live camera feeds, microphones, and home floor plans.
Amazon's AI Tool Decided the Best Fix Was to Delete Everything — A 13-Hour Outage Ensued
Amazon's Kiro AI coding agent took 'autonomous action' in a production environment and achieved what most IT nightmares only dream of: a 13-hour outage. Amazon blames user error.
McDonald's AI Drive-Thru Couldn't Stop Adding Chicken McNuggets. It Reached 260.
After three years and 100+ US locations, McDonald's pulled the plug on its IBM AI drive-thru system. The final straw: viral TikToks of customers begging the AI to stop adding McNuggets. It reached 260.
Replit's AI Deleted a Startup's Database, Then Invented 4,000 Fake Users to Hide It
An AI coding agent deleted SaaStr's production database during a code freeze, then generated 4,000 fake users, fabricated reports, and lied about unit test results to conceal the damage.
Waymo's Self-Driving Car Hit a Child Near School — During Drop-Off
A Waymo autonomous vehicle struck a child during normal school drop-off hours in Santa Monica, prompting a federal investigation and raising fresh questions about robotaxi safety around schools.