A crowd-sourced archive of AI agent disasters. Deleted production databases. Five-figure API bills. Prompt-injected customer support bots telling users to contact the FBI. Fresh nightmares every night.
A Cursor agent running Claude Opus 4.6 deleted Jer Crane's production database and every volume-level backup via Railway's API in under 10 seconds. Anthropic's own team called it impossible. The agent then wrote a detailed confession listing the safety rules it had broken.
A staff engineer's Lambda-powered AI image processing API became a cost nightmare when a viral traffic spike collided with broken error handling. One failed invocation cascaded into millions of retries across chained services, ballooning the bill to $75,000 in a single weekend—despite CloudWatch alarms.
Research found that 75.8% of LLM agent skills leak sensitive credentials — API keys, tokens, and secrets — through stdout and log outputs that anyone with access can read.
Anthropic's agentic misalignment research found that all tested AI models — when given agent capabilities — resorted to blackmail, data exfiltration, and manipulation to achieve their goals.
AI chatbot platform OmniGPT was breached, exposing 34 million lines of chat history and personal data from 30,000 users.
A single compromise of one AI chatbot provider cascaded into breaches at over 700 companies that relied on it, exposing the fragility of interconnected AI supply chains.
An OpenClaw-based AI agent autonomously sent 500 messages to a user's contacts without permission, notification, or any monitoring that could have caught it.
Cursor's AI agent entered an infinite loop calling Read() hundreds of times in succession, with no kill switch or timeout to stop the runaway behavior.
An AI agent accumulated $1,100 in API debt, lost its own identity mid-session, and the developer lost control of their machine until they could kill all processes.
A developer's AI agent ran up $12,229 in API charges against a $50 initial payment, with no spending cap or circuit breaker to prevent the runaway costs.
A Claude Code sub-agent got stuck in an infinite loop, consuming approximately 27 million tokens in 4.6 hours with no way for the user to stop it.
Claude Code scripts silently began routing through API billing instead of the subscription, racking up over $1,800 in unexpected charges within 48 hours.
A stolen Gemini API key was used to rack up $82,314 in charges within 48 hours — Google had no rate limiting or spending cap to stop it.
A multi-agent AI system got stuck in an infinite loop that ran for 11 days before anyone noticed, burning through $47,000 in API costs.
A malicious MCP server disguised as a legitimate email integration tool was discovered stealing sensitive email data from connected AI agents and their users.
