Rank #7 of 35·Data loss·26 Feb 2026·Claude Code
Claude Code runs terraform destroy, erasing 2.5 years of DataTalks.Club
“The cleanup was cleaner than expected. So was the database.”
- The real production state file was unpacked into the working folder
- The agent suggested terraform destroy; the founder agreed
- The DB and its snapshots were gone; AWS restored a hidden snapshot after 24h
He asked an AI to tidy up some cloud resources. It ran terraform destroy and took down 2.5 years of student work.
What happened
Alexey Grigorev, who runs DataTalks.Club, was moving another site onto the same AWS setup using Claude Code and Terraform, against the agent's own earlier advice. The Terraform state file was on his old laptop, so the agent began creating duplicate resources. While they cleaned those up, the real production state file got unpacked into the working folder. The agent suggested a full terraform destroy to save time, and he agreed.
The damage
The course platform's database, network, containers, load balancers and bastion host were all deleted. That included 2.5 years of homework, projects and leaderboards. The automated snapshots went with them, because they were tied to the database Terraform had just destroyed. Grigorev paid for a higher AWS support tier, and AWS restored a snapshot about 24 hours later. When he asked where the database had gone, he says the agent simply told him it was gone.
How it was caught, and what happened next
Grigorev knows Terraform well enough to teach it in his Zoomcamp courses, which have reached over 100,000 learners, and he still overruled the agent. Afterwards he added a nightly job that exports the latest backup to S3, kept separate from the database and the Terraform state, so even another destroy can't take it out.
Keep backups that live outside the infrastructure they protect, and never let an agent run destroy commands against production.
