---
id: "CMP-202611-hrios5"
page: "https://infrastack.ai/live/CMP-202611-hrios5"
kind: "campaign"
aliases: []
title: "Chinese-speaking attackers run ARTEX AI agents to steal AI model API keys and cloud logins from exposed LLM gateways and AI app builders (LiteLLM, One API, Dify, Langflow, RAGFlow, Open WebUI, Ollama) (Amazon)"
severity: "high"
cvss: null
epss: null
exploited: true
status: "open"
category: "ai-tooling"
products: [{"name": "LiteLLM", "fixed": [], "ecosystem": "pypi"}, {"name": "One API", "fixed": []}, {"name": "New API", "fixed": []}, {"name": "Veloera", "fixed": []}, {"name": "Sub2API", "fixed": []}, {"name": "Dify", "fixed": []}, {"name": "FastGPT", "fixed": []}, {"name": "Langflow", "fixed": [], "ecosystem": "pypi"}, {"name": "RAGFlow", "fixed": []}, {"name": "Open WebUI", "fixed": []}, {"name": "LobeChat", "fixed": []}, {"name": "Ollama", "fixed": []}]
ecosystems: []
tags: ["campaign", "early-warning"]
not_affected_if: []
published_at: "2026-11-13T00:00:00+00:00"
first_seen_at: "2026-11-16T07:54:01+00:00"
updated_at: "2026-11-16T07:54:25+00:00"
sources: [{"source": "analyst", "url": "https://aws.amazon.com/blogs/security/amazon-threat-intelligence-observes-threat-actors-running-cybercrime-campaigns-with-artex-ai-agents/", "first_seen_at": "2026-11-16T07:54:01+00:00", "cites": ["https://aws.amazon.com/blogs/security/amazon-threat-intelligence-observes-threat-actors-running-cybercrime-campaigns-with-artex-ai-agents/"]}]
exploit_sources: []
---

# Chinese-speaking attackers run ARTEX AI agents to steal AI model API keys and cloud logins from exposed LLM gateways and AI app builders (LiteLLM, One API, Dify, Langflow, RAGFlow, Open WebUI, Ollama) (Amazon)

Link for people: https://infrastack.ai/live/CMP-202611-hrios5 (this .md is for agents)

Amazon Threat Intelligence (13 Nov 2026): from 3 Oct to 5 Nov 2026 its MadPot decoys posing as LLM inference APIs caught ARTEX agents sending their own campaign memory. Chinese-nexus groups (moderate confidence) ran ARTEX agents on near-autopilot to find internet-exposed AI apps, pull the upstream provider API keys they store, test them and pool them; the stolen keys paid for the LLM inference driving further hunting. Cloud credentials were used to enumerate storage, managed AI model services, secrets and parameter stores. Targets: LLM gateways/proxies One API, New API, Veloera, Sub2API, LiteLLM; AI app builders Dify, FastGPT, Langflow, RAGFlow, Open WebUI, LobeChat, Ollama. Routes: unclaimed uninitialized installs, anonymous management and provider-configuration endpoints, default admin credentials (LiteLLM default master key), CVE-2025-3248 (Langflow), CVE-2026-28797 (RAGFlow), an isCliRequest header auth bypass in One API family systems, SSRF to cloud metadata, and secrets in client-side JavaScript, source maps, debug routes and env files. Keys were also harvested from GitHub repos and event archives, npm and PyPI tarballs, container image layers, notebooks, model-hosting spaces, paste sites and Telegram. Amazon notified affected AWS customers; which other organizations were hit is unknown, and outcomes the agents reported are not verified. A second campaign tried deepfake KYC bypass at two crypto exchanges (all recorded attempts rejected).

_Summary, check and fix written by a model from the sources; labelled as such._

## Affected

- pypi:LiteLLM affected range not listed fixed no fixed version listed
- :One API affected range not listed fixed no fixed version listed
- :New API affected range not listed fixed no fixed version listed
- :Veloera affected range not listed fixed no fixed version listed
- :Sub2API affected range not listed fixed no fixed version listed
- :Dify affected range not listed fixed no fixed version listed
- :FastGPT affected range not listed fixed no fixed version listed
- pypi:Langflow affected range not listed fixed no fixed version listed
- :RAGFlow affected range not listed fixed no fixed version listed
- :Open WebUI affected range not listed fixed no fixed version listed
- :LobeChat affected range not listed fixed no fixed version listed
- :Ollama affected range not listed fixed no fixed version listed

## Cause

How the attacker got in: not disclosed yet (unknown). The record lists many routes (unclaimed installs, anonymous endpoints, default credentials, CVEs, header bypass, SSRF, exposed secrets) without saying which one any affected victim used, while keys stolen from victims were then used by the attackers to pay for further hunting.

How it reached others: credentials stolen there were used against others (stolen-credential).

Last month's incidents with the same cause, and the full checklist: https://infrastack.ai/live/breaches.md#stolen-credential

## How to check

Do you run any of LiteLLM, One API, New API, Veloera, Sub2API, Dify, FastGPT, Langflow, RAGFlow, Open WebUI, LobeChat or Ollama reachable from the internet? Check for: setup/init pages still claimable; admin or provider-config endpoints answering without login; LiteLLM still on the default master key; Langflow unpatched for CVE-2025-3248; RAGFlow unpatched for CVE-2026-28797; One API-family requests carrying an isCliRequest header. Search access logs for the Campaign 1 IPs 104.28.225[.]223, 31.22.91[.]156, 150.136.38[.]110, 128.242.115[.]131, 216.176.239[.]231 (seen 14 Oct - 2 Nov 2026). Watch AI provider billing for unexpected spend or usage from unknown IPs. Look for provider keys in public repos, published npm/PyPI packages, container image layers, front-end JS and source maps.

## Fix

Take AI gateways and AI app builders off the public internet or put them behind SSO/VPN; finish setup on any uninitialized install; change default admin credentials and the LiteLLM master key; patch Langflow (CVE-2025-3248) and RAGFlow (CVE-2026-28797); block SSRF to cloud metadata (enforce IMDSv2). If an instance was exposed, rotate every upstream provider API key and cloud credential it holds, then review provider usage and cloud logs (identity lookups, storage listing, secrets/parameter store reads, managed model calls). Block the listed IPs.

## Timeline

- 2026-11-16T07:54:01+00:00 analyst https://aws.amazon.com/blogs/security/amazon-threat-intelligence-observes-threat-actors-running-cybercrime-campaigns-with-artex-ai-agents/

## Documents

- [describes] Amazon Threat Intelligence observes threat actors running cybercrime campaigns with ARTEX AI agents https://aws.amazon.com/blogs/security/amazon-threat-intelligence-observes-threat-actors-running-cybercrime-campaigns-with-artex-ai-agents/
- [describes] @TheHackersNews: 🚨 Amazon uncovers AI agents hunting API keys across LiteLLM, Dify, and other AI  https://x.com/TheHackersNews/status/2122232175529766932
- [describes] @TheHackersNews: One exposed AI gateway. Multiple provider API keys at risk.

Amazon's captured l https://x.com/TheHackersNews/status/2122232821351890997
- [describes] @TheHackersNews: The scary part? These attacks don't need new exploits.

Amazon found AI agents t https://x.com/TheHackersNews/status/2122232893032579394

---

_Free to read, with fair use of about 60 pages a minute. For higher limits or API access: sales@infrastack.ai_
