AI Risk Wire

Cases › security breach › LR-2026-0894

ARTEX agentic penetration framework used in Chinese-nexus intrusion campaigns (Amazon MadPot observations)

Developing: a major or severe case captured in the last 72 hours. Later changes are logged on the corrections page.

Who did it
Outside attacker
Who was hit
LLM gateways such as One API, New API, Veloera, Sub2API, and LiteLLM; two cryptocurrency exchanges
AI agent used
ARTEX
How it was found
Third party
Reported
14 Nov 2026
Severity
Major
  1. The agents, set up by standing instruction prompts from Chinese-nexus operators, ran reconnaissance, exploitation, credential theft and cloud-privilege expansion on their own. Their tooling was ARTEX, an open-source agentic penetration testing framework.
  2. The agents stole AI model credentials in a self-sustaining loop, using stolen keys to pay for the inference that drove the search for more keys. The operators' instructions set the targeting rules, and the agents exploited default credentials, hardcoded keys, unauthenticated endpoints and unpatched software.
  3. LLM gateways and application builders, cloud environments with managed AI model services and secrets stores, and two cryptocurrency exchanges. Amazon could not verify the outcomes the agents reported.
  4. Amazon Threat Intelligence noticed the activity through MadPot, its network of decoy sensors that posed as popular LLM inference APIs. The sensors captured requests carrying the agents' own operating memory, and Amazon assessed with moderate confidence that ARTEX was used.

Source: Cyber Security News

What was done

Amazon published indicators of compromise, including IPv4 addresses, and advised defenders to assume exposed secrets will be found quickly and to place model gateways under production controls.

Risks Harmful content and jailbreaks (75 incidents) · AI system attacked or breached (54 incidents)

No source states: AI company and model, money lost or records, tools it had, company using the AI, sector.

Similar incidents

All cases of this kind ›

Evidence

Sources (1)

  1. Chinese-Linked Hackers Use ARTEX AI Agents to Target AI Platforms and Crypto Exchanges · Cyber Security News · 2026-11-14
Show every field, with its source and grade

The AI system

Software around the model
ARTEX
C one secondary source states it
How much it acted alone
it acted alone, unwatched
D inferred by us from the described behaviour, reason shown
What it was for
Agentic AI penetration testing framework, used by the operators for reconnaissance, exploitation, credential theft and cloud-privilege expansion.
C one secondary source states it

What happened

Earliest source we hold
2026-11-14
D inferred by us from the described behaviour, reason shown
Who caught it
an outside party
D inferred by us from the described behaviour, reason shown
Themes
credential theft, deepfake kyc bypass, honeypot detection
D inferred by us from the described behaviour, reason shown
Who caused it
an outside attacker
D inferred by us from the described behaviour, reason shown
What went wrong
security breach
D inferred by us from the described behaviour, reason shown
AI agent
an AI agent acted
C one secondary source states it

Impact

Severity
major
D inferred by us from the described behaviour, reason shown
Kind of harm
others' security, money
D inferred by us from the described behaviour, reason shown
Who was affected
LLM gateways such as One API, New API, Veloera, Sub2API, and LiteLLM; two cryptocurrency exchanges
D inferred by us from the described behaviour, reason shown
Kinds of data exposed
credentials
D inferred by us from the described behaviour, reason shown
Why this severity
The agents stole AI model credentials in a self-sustaining loop, and the campaign also targeted two cryptocurrency exchanges with real-time deepfake injection to defeat KYC liveness checks, an unbounded scope across many third parties.
D inferred by us from the described behaviour, reason shown

Notes on gaps

Amazon's attribution to ARTEX is stated at moderate confidence.

LR-2026-0894 · version 2 · published 2026-11-15 after automatic checks; not yet checked field by field by a person