Safety failure
Realized harm
A Claude Code agent ran terraform destroy and wiped DataTalks.Club's production infrastructure, database and snapshots
AIR-2026-0062 AIR record · 2026-02-claude-code-terraform-destroy-datatalks
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
~11:00 PM: A Terraform auto-approve command inadvertently wiped out all production infrastructure, including the Amazon Relational Database Service (RDS). I later discovered that all snapshots were also deleted, prompting me to create an AWS support ticket.alexeyondata.substack.com
The record
Disclosed
2026-03-06
Occurred
2026-02-26
Vendor
Anthropic
Product
Claude Code
Risk domain
Agent behaviour › Unsafe actuation
System type
Generative model
Who acted
Agent acted — the agent ran terraform destroy against production
Attack surface
Coding agent
Autonomy level
Semi-autonomous
Initial vector
None adversarial
Impact type
Data destruction, Business disruption
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI01 – Agent Goal Hijack
- MITRE ATLAS
- AML.T0048 – External Harms
- NIST AI RMF
- MEASURE 2.6; MANAGE 1.3
- EU AI Act
- Annex III §1(c)
- ISO/IEC
- TR 24028; 24029-1; 23894
- AIUC-1
- C. Safety
- Scope
- agent misuse
- Mitigation
- Require confirmation for destructive actions; Implement dry-run modes; Use layered safety controls.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
VPC, ECS cluster, load balancers, bastion host, RDS database and automated snapshots; 2.5 years of course data at risk
Capability abused
Defense & remediation
Recorded defense
none - auto-approve was enabled and deletion protection was not
Guardrail kinds
None
Guardrail outcome
Not applicable
Fix shipped
AWS restored a snapshot; owner added deletion protection, S3 backups and remote state
Taxonomy & references
Curator notes
AI Incident Database incident 1424; first-person account by Alexey Grigorev. Root enabler is IaC plus auto-approve: a single agent action collapses an entire environment.
Supporting sources
Cite as
Enkrypt AI. "A Claude Code agent ran terraform destroy and wiped DataTalks.Club's production infrastructure, database and snapshots". AIR, AIR-2026-0062. https://agentincident.vercel.app/i/AIR-2026-0062/ (accessed 2026-10-19).