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

LLMjacking: stolen cloud credentials resold for access to hosted LLM services

AIR-2024-0018 AIR record · 2024-05-sysdig-llmjacking
Open in dashboard About
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
the cost to the victim can be over $46,000 per daysysdig.com
The record
Disclosed
2024-05-06
Occurred
2024
Vendor
multiple
Product
AWS Bedrock, Anthropic, OpenAI, Azure OpenAI and other hosted model endpoints
Risk domain
Access control and permissions › Credential theft
Also touches
Unsafe actuation
System type
Generative model
Who acted
AI was the target (contested) — cloud credentials stolen through a Laravel CVE; the LLM service is the abused resource
Attack surface
Agent framework
Autonomy level
Assistant
Initial vector
Credential theft via cve 2021 3129 laravel
Impact type
Service abuse, Financial loss, Credential theft
Confidence
High
Submitted by
EAI Internal
Crosswalk
OWASP Agentic Top 10 (2026)
ASI03 – Identity and Privilege Abuse
MITRE ATLAS
AML.T0012 – Valid Accounts
NIST AI RMF
MEASURE 2.7
EU AI Act
Article 13
ISO/IEC
TR 24028; 42001; 23894
AIUC-1
B. Security
Scope
tool failure
Mitigation
Use secure credential storage; Implement short-lived tokens.
Taxonomy
Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
Ten AI service providers affected; modelled cost to a single victim over $46,000 per day
Capability abused
model_inference_api_accessquota_consumption
Taxonomy & references
OWASP ASI03 Identity and Privilege AbuseCVE-2021-3129llmjackingcredential_theftcost_shifting
Curator notes
Origin point for the whole LLMjacking category that later shows up as reverse-proxy resale, Xanthorox-style criminal AI services and GTIG's account-pooling clusters. The $46,000/day figure is Sysdig's modelled worst case at maximum quota, not an observed bill. Attackers checked logging configuration before use, showing deliberate evasion.
Supporting sources
Cite as
Enkrypt AI. "LLMjacking: stolen cloud credentials resold for access to hosted LLM services". AIR, AIR-2024-0018. https://agentincident.vercel.app/i/AIR-2024-0018/ (accessed 2026-10-19).