Claude Code AI agent destroys production DB via Terraform command

Fri, 10 Jul 2026 11:00:57 -0500

Platform/SRE — Plan: This incident—an AI coding assistant given unconstrained Terraform access wiping a production database—is a concrete signal to audit and restrict AI agent permissions to production IaC state; plan to implement plan-before-apply gates, workspace isolation, and state-level protections before allowing any AI assistant to execute Terraform in production environments.
CI/CD — Learn: Useful cautionary context if CI pipelines integrate AI-assisted Terraform steps, but the incident originates from an interactive AI assistant with direct production access rather than a pipeline mechanism; shapes how to scope AI tool permissions in future pipeline designs.
Leader — Plan: This high-profile incident—145 upvotes, 158 comments—is a concrete risk signal for any org adopting AI coding assistants; evaluate and formalize org-wide policy on AI agent access to production systems, and mandate guardrails (dry-run gates, least-privilege IAM, human approval for destructive operations) as a standard before broader rollout.

Incident-Report on CuraDevOps