---
title: Amazon's LLM honeypots caught ARTEX agents' full campaign memory: a self-funding AI-key theft operation and deepfake attacks on crypto-exchange KYC | ctxsec
url: https://ctxsec.com/items/amazon-madpot-artex-agents-ai-credential-theft-self-funding-kyc-deepfake-bypass
description: "From Oct 3 to Nov 5, Amazon's MadPot sensors posed as LLM inference APIs, and ARTEX agents sent them requests carrying their system prompts, tool history and shared knowledge graph. Amazon assesses with moderate confidence these came from disparate Chinese-nexus groups. The largest campaign stole AI-provider keys from exposed LLM gateways and AI app builders, validated them, and fed them into the framework's own model gateway to pay for more hunting; operators graded keys L1 to L3 and treated keys public for over 30 days as low value. A second campaign tried to beat two crypto exchanges' identity checks with real-time deepfakes and doctored ID photos; the agents' best attempt matched the liveness movement sequence within 7 pixels, but the vendor's server-side check rejected every recorded try."
status: "vendor research (Amazon Threat Intelligence, Nov 13, 2026); outcomes come from the agents' own records and Amazon says it can't verify them"
published: 2026-11-13
---

> From Oct 3 to Nov 5, Amazon's MadPot sensors posed as LLM inference APIs, and ARTEX agents sent them requests carrying their system prompts, tool history and shared knowledge graph. Amazon assesses with moderate confidence these came from disparate Chinese-nexus groups. The largest campaign stole AI-provider keys from exposed LLM gateways and AI app builders, validated them, and fed them into the framework's own model gateway to pay for more hunting; operators graded keys L1 to L3 and treated keys public for over 30 days as low value. A second campaign tried to beat two crypto exchanges' identity checks with real-time deepfakes and doctored ID photos; the agents' best attempt matched the liveness movement sequence within 7 pixels, but the vendor's server-side check rejected every recorded try.

[← AI](/ai) · [markdown](/items/amazon-madpot-artex-agents-ai-credential-theft-self-funding-kyc-deepfake-bypass.md)

vendor research (Amazon Threat Intelligence, Nov 13, 2026); outcomes come from the agents' own records and Amazon says it can't verify them

2026-11-13 7:41 PM ET

### Amazon's LLM honeypots caught ARTEX agents' full campaign memory: a self-funding AI-key theft operation and deepfake attacks on crypto-exchange KYC

- **Why it matters:** nothing here is a new technique; the agents won where keys sat in code and artifacts, interfaces had no auth, and defaults were never changed. Speed and persistence are the new part.
- **Do:** put AI gateways and inference endpoints under production controls (auth, patching, no default master keys), scan public artifacts for provider keys and revoke fast, and alert on unexpected model spend.
- **Signal:** at least four ARTEX sessions ran at once on the KYC goal, handing off tasks through cross-session tools.
- IOCs are in the ARTEX hunt item.

Sources: [AWS Security Blog: Amazon Threat Intelligence observes threat actors running cybercrime campaigns with ARTEX AI agents (Nov 13, 2026)](https://aws.amazon.com/blogs/security/amazon-threat-intelligence-observes-threat-actors-running-cybercrime-campaigns-with-artex-ai-agents/)
