▶ Watch on YouTube
Technology

AI agents steal keys to artificial intelligence services, Amazon warns

2026-11-16 · Nuevo Día
AI agents steal keys to artificial intelligence services, Amazon warns

Autonomous artificial intelligence agents have been hunting for access keys to AI services in internet-exposed applications, Amazon's threat intelligence team revealed on Monday.

The agents targeted applications such as LiteLLM, One API and Dify, which store AI provider keys, and exploited default administrator credentials, unfinished installations and known flaws to steal them. The stolen keys were then used to pay for the models that powered the hunt for more keys, in a self-sustaining operation.

Amazon detected the activity between October 3 and November 5 through its MadPot network of decoy servers. The company warns it could not verify everything the agents claimed to have achieved, but it could see what they were after. With moderate confidence, it attributes the campaign to separate China-linked groups, citing Simplified Chinese prompts and the time zone of the operations.

In a separate case, Cloudflare researchers saw the ARTEX tool, an agentic AI penetration-testing program, gain access to an account token with broad permissions and generate more than 400 database dump requests in an hour.

Amazon recommends changing default credentials, not leaving unfinished installations exposed on the internet, and watching for unexpected spending on AI accounts. The FBI also warned this week about attackers using ARTEX since at least August.

Photo: The Hacker News

Source: The Hacker News

← More news in Technology · Leer en español