Amazon says stolen AI API keys power agents hunting for more

For the CISO of a regional US bank

Amazon Threat Intelligence said in a blog post that its MadPot decoy sensors captured records between Oct. 3 and Nov. 5 showing AI agents running on the open-source ARTEX framework stole AI provider keys and used them to pay for model calls that searched for more keys.2

The records came from agents that sent requests to MadPot sensors set up to look like LLM inference APIs, and those requests carried the agents' own campaign histories, according to Amazon.2 Amazon said it assesses with moderate confidence that ARTEX was used, based on prompt references to ARTEX tooling and other technical artifacts, and that it did not observe the activity as it happened and cannot verify the outcomes the agents reported.2

According to Amazon, the agents searched exposed LLM gateways and AI application builders including LiteLLM, Dify and Langflow, validated the provider keys they found and stored working ones in a managed pool; the playbook included default admin credentials such as LiteLLM's default master key and CVE-2025-3248 in Langflow.21

The FBI cited ARTEX-related attacks on South Korean financial institutions that resulted in the theft of personal information belonging to around 67,000 people, according to VentureBeat's account of the FBI's Nov. 13 advisory; The Register dates those attacks to August.13

CrowdStrike's Adam Meyers told VentureBeat that human operators throw "dozens of commands an hour at a target," while "the most we've seen was like 1,500 commands in an hour" with an agent.1

The FBI said some ARTEX intrusions progressed from initial targeting to privileged access in under three hours, and that ARTEX's developer removed the original GitHub repository in October, while forks remain and Censys identified 1,465 hosts running an exposed ARTEX console as of Nov. 4, according to The Register.13

Why it matters

Any LLM gateway or AI app builder a bank team stood up for a pilot may hold live provider keys and sit on the internet with default settings, so it needs a named owner, an inventory entry and a credential review. Unexpected model-provider charges or unfamiliar LLM proxies are now a security signal as well as a cost issue.

From the brief

Our take. Your developers' AI gateway is a drawer of model-provider keys, and attackers now spend those keys to run agents that find the next drawer. The FBI saw some intrusions reach privileged access in under three hours.

Action. Security engineering: Inventory internet-facing LiteLLM, Dify and Langflow installs, rotate default master keys, and patch Langflow CVE-2025-3248.

Sources

  1. ReportAmazon says stolen AI API keys power agents hunting for more Venturebeat
  2. Primary sourceAmazon Threat Intelligence observes threat actors running cybercrime campaigns with ARTEX AI agents | Amazon Web Services Amazon
  3. Also reported byAttackers are using AI agents to automate cybercrime at 'industrial scale,' FBI and AWS warn Theregister