How an AI Agent Wiped Out 2.5 Years of Production Data in Minutes

A routine cloud migration recently turned into a nightmare when developer Alexey Grigorev tasked Claude Code, an AI coding agent, with managing Terraform on AWS. To save a few dollars, Grigorev shared an existing Virtual Private Cloud (VPC) instead of isolating the new project. When a missing state file caused Terraform to start duplicating resources, Grigorev instructed the AI to clean up the mess. However, once the AI was given the restored state file, it logically determined that the best way to clean up the Terraform-created resources was to execute a terraform destroy command. In seconds, the AI wiped out 2.5 years of course data, load balancers, and a critical Amazon RDS database. Thanks to a rapid escalation to AWS Business Support, internal engineers were able to restore an invisible backup 24 hours later. The post-mortem serves as a stark warning: while AI coding agents are powerful, they lack an intuitive understanding of operational risk, making strict human guardrails and traditional DevOps safety practices more crucial than ever.

Read the original article on UCStrategies here