AI Agents Automate Theft of API Keys from Exposed Apps
Amazon and Cloudflare report that AI agents are systematically targeting internet-exposed applications to steal AI provider keys.
Amazon’s threat intelligence team identified a campaign using AI agents to locate weak spots in internet-exposed applications such as LiteLLM, One API, and Dify. The agents extracted stored AI provider keys and used them to run models that powered the search for additional credentials. Captured records include a playbook listing default administrator credentials and known flaws in targeted products. Amazon stated it has moderate confidence the agents ran on the ARTEX framework, though it cannot verify the specific outcomes reported by the agents.
Cloudflare incident responders observed ARTEX gaining access to a compromised API token used for CI/CD without IP restrictions or rate limiting. This access led to 400 database dump requests within an hour. The system operates with an operator providing targets, a planner agent splitting goals into tasks, and worker agents executing security tools. Decoy servers named MadPot posed as popular LLM services and received requests from October 3 to late last week.
Amazon assessed with moderate confidence that the operators were separate China-linked groups. Microsoft reported last week that attackers were using agentic AI tools for GitHub secret hunting. The FBI issued a FLASH alert this week stating that a range of attackers have been active.
Sources
- The Hacker NewsAmazon Uncovers AI Agent Campaign Targeting LiteLLM, Dify, and Other Apps for API Keys ↗
- The RegisterAttackers are using AI agents to automate cybercrime at 'industrial scale,' FBI and AWS warn ↗
Every sentence of this story was checked against facts quoted from these articles. How we make the news
Spotted an error? Tell us
