Agent Incident RegistryAIR
v1 · updated 19 Oct 2026
In the wild Realized harm

GTG-5004: actor with no independent malware skill used Claude to build and sell ransomware variants

AIR-2025-0122 AIR record · 2025-08-anthropic-gtg-5004-ransomware-raas
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
used Claude to develop, market, and distribute several variants of ransomwareanthropic.com
The record
Disclosed
2025-08-27
Occurred
2025
Vendor
Anthropic
Product
Claude
Risk domain
Agent behaviour › Unsafe actuation
System type
Generative model
Who acted
Human acted on output — Claude wrote the ransomware; the actor sold and buyers deployed it
Attack surface
Coding agent
Autonomy level
Assistant
Initial vector
Direct misuse for malware development
Impact type
Malware distribution, Commercial sale of malware
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
Ransomware sold on dark web forums for $400-$1200 per variant
Capability abused
code_generationdebuggingevasion_technique_design
Defense & remediation
Recorded defense
Claude safety training
Guardrail kinds
Abuse detection
Guardrail outcome
Silent
Fix shipped
Accounts banned (Aug 2025)
Taxonomy & references
OWASP ASI02 Tool Misuse and Exploitationcapability_upliftraasmalware_development
Curator notes
Anthropic's assessment that the actor 'could not implement or troubleshoot core malware components' without Claude is the load-bearing claim about uplift; it is a vendor inference from prompt history, not externally verified. Downstream victim impact from the sold ransomware was not quantified.
Cite as
Enkrypt AI. "GTG-5004: actor with no independent malware skill used Claude to build and sell ransomware variants". AIR, AIR-2025-0122. https://agentincident.vercel.app/i/AIR-2025-0122/ (accessed 2026-10-19).