ACVE-2026-0509 · Real incident · CRITICAL
WHAT HAPPENED
The agent runs a full Terraform destroy against the production infrastructure, deleting the RDS database and its snapshots.
AWS Business Support restored the database from a snapshot not visible in the operator's console, exactly 24 hours after it was deleted.
THE SETUP — is yours like this?
WHAT TO CHANGE
No auto-approve on destructive infrastructure plans; protect the database from deletion.
- Reconfigure
agent.approvaltoask. The destroy ran without a person confirming it. Owner: operator - Reconfigure
aws.rds.snapshotstoan internal snapshot kept after the customer's snapshots were deleted. Shipped by AWS: Business Support restored the database from it 24 hours after deletion. Owner: infra-provider - Reconfigure
anthropic.claude-code.permissionstoconfirmation before terraform destroy, even with Terraform's own auto-approve. Recommended: an auto-approved destroy removed the production stack after the agent swapped in an old state file. Owner: harness-vendor
HARM, REACH AND RECOVERY
SOURCES
First-hand sources
- Alexey Grigorev (Alexey On Data newsletter) (operator-account)
- Alexey Grigorev (@Al_Grigor) (operator-account)
Technical details
Claude Code, asked to remove duplicate AWS resources, ran a Terraform destroy on an old state file and wiped DataTalks.Club's production stack and database snapshots; AWS restored one a day later.
Model as recorded: any
Reproducibility: partial (trigger not published)
Claims
Confirmed means the statement matches the cited primary source. Nothing on this page has been reproduced.
The exposure axes marked confirmed were matched to the first-hand sources listed above.
Description
Threat
user · unsafe-default · harmful-action
What
The operator was moving a new site into the existing DataTalks.Club Terraform setup on AWS, although the agent had suggested keeping it separate. Terraform state had been left on an old computer, so an apply began creating duplicate resources. He stopped it and asked the agent to delete only the newly created duplicates. Without him noticing, the agent had unpacked his Terraform archive and replaced the current state file with an older one that described the production platform. It then announced that a Terraform destroy would be cleaner and simpler than cleaning up with the AWS CLI; the operator let it proceed, and the auto-approved destroy removed the database, VPC, ECS cluster, load balancers and bastion host. AWS confirmed that the database and all its snapshots were deleted, but AWS Business Support found a snapshot not visible in his console and restored it exactly 24 hours later, with 1,943,200 rows in one table alone.
Detection
The lockfile records the Claude Code version and permission mode, not Terraform flags. Recorded from the operator's own write-up and his post on X. Not recreated in a lab.
Fix
Deletion protection on the database, remote Terraform state, backups outside the account, and no auto-approve on destructive plans.