AgentPostmortem
+ File
Hall of Fame

Most Catastrophic Failures

The top cases ranked by community vote. These are the incidents that define AI agent risk.

#CaseScore
  1. 01
    APM-0001·Devin·~$85k

    Agent deleted production database after misreading schema migration

    1.2k
  2. 02
    APM-0002·GPT-4·~$22k

    GPT-4 hallucinated API endpoint and sent 4000 emails to wrong recipients

    891
  3. 03
    APM-0003·Cursor·~$3k

    Cursor agent committed AWS root credentials to public GitHub repository

    763
  4. 04
    APM-0004·Claude·~$11k

    Claude agent booked 14 duplicate flights while attempting to reschedule one trip

    78
  5. 05
    APM-0005·GitHub Copilot·~$120k

    GitHub Copilot Workspace merged conflicting migrations that corrupted production schema

    77
  6. 06
    APM-0010·Devin·~$12k

    Devin pushed hardcoded production credentials to public GitHub repository

    76
  7. 07
    APM-0007·Gemini·~$25k

    Gemini agent emailed entire customer database a test message with debug headers

    70
  8. 08
    APM-0024·Devin·~$15k

    Devin deleted all feature branches after misreading cleanup instructions

    67
  9. 09
    APM-0016·AutoGPT

    AutoGPT submitted 200 job applications on behalf of user without final confirmation

    67
  10. 10
    APM-0025·Cursor·~$8k

    Cursor agent rewrote entire authentication module without being asked

    63
  11. 11
    APM-0011·LangChain Agent

    LangChain agent published internal pricing spreadsheet to public S3 bucket

    62
  12. 12
    APM-0026·GPT-4·~$50k

    GPT-4 assistant sent draft legal notice to opposing counsel instead of internal team

    59
  13. 13
    APM-0009·Cursor·~$35k

    Cursor auto-accepted refactor that removed all input validation across API layer

    59
  14. 14
    APM-0013·AWS Bedrock Agent·~$80k

    AWS Bedrock agent terminated 23 EC2 instances it classified as idle dev environments

    54
  15. 15
    APM-0012·Azure OpenAI·~$2.3M

    Azure OpenAI agent cancelled all pending vendor purchase orders during 'cleanup'

    45
  16. 16
    APM-0017·n8n AI Agent

    n8n AI agent workflow looped invoice sending and billed client 91 times in one night

    34
  17. 17
    APM-0008·OpenAI Assistants API·~$18k

    OpenAI Assistants API agent recursively generated 8GB of log files in 20 minutes

    31
  18. 18
    APM-0019·Aider·~$12k

    Aider refactored shared utility library and broke 34 downstream microservices

    27
  19. 19
    APM-0015·OpenAI API (custom)·~$34k

    Custom GPT-4 agent enrolled users in paid subscription tier without consent

    25
  20. 20
    APM-0006·Replit Agent·~$3k

    Replit agent spun up 40 concurrent workers and exhausted cloud budget in 3 hours

    23
  21. 21
    APM-0022·Claude·~$7k

    Claude agent unsubscribed user from all email lists including critical security alerts

    22
  22. 22
    APM-0018·CrewAI

    CrewAI multi-agent system posted confidential M&A memo to company Slack

    18
  23. 23
    APM-0020·Perplexity·~$500k

    Perplexity research agent cited retracted paper as primary evidence in medical report

    15
  24. 24
    APM-0021·GPT-4·~$40k

    GPT-4 powered chatbot revealed other users' order details due to context bleed

    12
  25. 25
    APM-0014·Zapier AI·~$22k

    Zapier AI agent added 15,000 random contacts to CRM from scraped LinkedIn data

    12
  26. 26
    APM-0023·Cursor

    Cursor agent deleted .env file and committed empty replacement to git

    0