A developer over-relied on Claude Code AI agent to execute Terraform infrastructure commands, which accidentally destroyed production infrastructure for DataTalks.Club course management platform, wiping 2.5 years of student submission data and backups.
Alexey Grigorev was migrating his AI Shipping Labs website from GitHub Pages to AWS and planned to share infrastructure with his existing DataTalks.Club course management platform to save costs. He used Claude Code AI agent to execute Terraform infrastructure management commands but forgot to upload the state file from his old computer. When Claude ran 'terraform plan', it assumed no existing infrastructure existed and began creating duplicate resources. Grigorev stopped the process and had Claude analyze and delete the duplicates using AWS CLI. Later, Claude unpacked an archived Terraform state file and ran 'terraform destroy' to clean up what it believed were temporary resources. However, this actually destroyed the entire production infrastructure for DataTalks.Club, including the RDS database containing 2.5 years of course data with 1,943,200 rows in the courses_answer table alone, plus all automated snapshots. The incident occurred on February 26th around 11 PM. Grigorev upgraded to AWS Business Support for faster response times and after 24 hours, AWS support was able to restore the database from a snapshot that was not visible in his console. He implemented multiple safeguards including deletion protection, separate backups, daily restore tests, and removing Claude's ability to execute Terraform commands automatically.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that fail to perform reliably or effectively under varying conditions, exposing them to errors and failures that can have significant consequences, especially in critical applications or areas that require moral reasoning.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
