Amazon’s threat intelligence team says AI agents have been hunting internet-exposed AI apps such as LiteLLM, Dify, One API, and similar tools for the provider API keys stored inside, The Hacker News reports, drawing on Amazon’s own write-up. The team captured the agents’ records with decoy servers in its MadPot network, from October 3 through early November. Stolen keys, the records say, were used to run the models that powered the search for more keys. Amazon has moderate confidence the agents ran on ARTEX, the open-source AI pentest framework. It also says it cannot verify all the outcomes the agents reported, because that activity happened on systems the decoys do not watch.
The practical advice is straightforward. Operators should change default administrator credentials, not leave unfinished installs on the public internet, apply vendor updates, and rotate credentials if they think an install was exposed. Amazon notified AWS customers who may have been affected. I wrote earlier about the FBI’s FLASH warning on ARTEX, and about a separate LiteLLM supply-chain incident. This is a different problem: exposed installs and weak admin setups, not a trojaned package.
My take
If you stood up a LiteLLM, Dify, or similar gateway and left it on the internet with default credentials, treat that as urgent. Change the defaults, patch it, and assume any stored keys need a fresh rotation. AI apps that hold a lot of provider keys are a juicy target, and you should not be surprised when automated agents start knocking.
God Bless, and Happy hacking!
References & Further Reading
- Amazon uncovers an AI-agent campaign against exposed AI apps (The Hacker News)
- Amazon Threat Intelligence on cybercrime campaigns with ARTEX AI agents (AWS Security Blog)
- Update: the FBI issues a FLASH warning on ARTEX (EtherNinja)
- Update: the LiteLLM supply-chain attack started with an exposed master key (EtherNinja)