AI agents stole API keys to bankroll a month of autonomous cybercrime, Amazon warns
In an unusual twist, Amazon’s account draws on the AI agents’ own descriptions of their operations.

AI agent is a hacker's best new friend. By Cybernews.
- Amazon says attackers used Artex AI agents to steal API keys from exposed AI apps for about a month.
- The agents verified stolen keys and pooled them to pay for more large language model requests.
- Not every listed target was necessarily breached, but affected organizations should investigate quickly.
- Amazon links the activity with moderate confidence to threat groups connected to China.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
Attackers’ AI agents spent a month stealing API keys from web-exposed AI apps such as LiteLLM or Dify to keep funding their credential hunting campaigns, Amazon has warned. The agents ran on Artex, an open-source AI tool for penetration testing.
The campaign began on October 3rd and was, according to Amazon, disrupted on November 5th when the company’s MadPot, a global network of decoy threat sensors, captured requests from AI agents containing their operating memory.
This gave the cyber investigators a view into how the discovered campaigns were being run.
It also means that the AI agents helped expose their own activities. They sent requests to decoy systems configured to look like popular services for accessing large language models, and those requests carried records of their offensive campaigns.
Stay updated with our latest stories and follow us on social media
Be the first to discover new stories, ideas, and updates from our team.
The latter were run on Artex, an open-source agentic AI penetration testing framework, Amazon said on their security blog before revealing it had notified potentially affected Amazon Web Services (AWS) customers.
Secrets will be found quickly
Attackers can evidently repurpose these security checks for criminal operations. In the main campaign, the agents hunted for internet-exposed AI apps, extracted the API keys, verified that they worked, and added them to a shared pool.
According to Amazon, hundreds of targets named in the AI agents’ records include LLM gateways and proxies such as LiteLLM, One API, or Veloera, and AI app builders such as FastGPT or Dify.
Curiously, “the most prominent Artex campaign was a large-scale theft of credentials to access AI models, organized as a self-sustaining operation in which stolen keys paid for the LLM inference powering further credential hunting,” said the company.
Another campaign targeted cryptocurrency exchanges for know-your-customer (KYC) verification bypass using real-time deepfake injection.
One important caveat is that not every target mentioned in the discovered records was necessarily breached. Amazon stressed it was publishing its report to help organizations investigate themselves.
Still, the findings seem worrying – especially to firms building AI features into websites, internal tools, and customer-facing applications.
Given the speed and persistence of the attackers, organizations should assume exposed secrets and misconfigurations will be found quickly and remediate accordingly,Amazon warned.
Plus, the way the attackers operated looks very methodical. For example, in one credential theft campaign, operators told the agents to work through a fixed sequence of countries, prioritize enterprise production sites over personal projects, and grade every stolen credential by whether it worked and what it could reach.
Artex is powering global cyberattacks
Amazon Threat Intelligence has assessed “with moderate confidence” that the actors behind the campaigns were disparate Chinese-nexus threat groups, citing the use of simplified Chinese language in prompts, China time-zone configurations, targeting consistent with Chinese interests, and tooling predominantly used by Chinese-nexus threats.
Besides, sample instructions included setting red lines, such as excluding mainland China targets and government domains, and required all outbound traffic to go through proxies.
Other tech companies have recently also named Artex as an attack vector in credential hunting campaigns. They also see Chinese traces in those attacks.
Has your password leaked?
For example, Microsoft warned in late October that Storm-0940, a China-state aligned actor, used Artex to automate attacks targeting hundreds of organizations globally across a broad range of industries.
Soon after, Google Threat Intelligence Group (GTIG) also said it had observed threat groups weaponizing agentic frameworks to execute highly automated intrusions.
“Security teams, developers, and system administrators will need to fundamentally rethink defense strategies to counter scalable, autonomous software,” said GTIG.
Finally, the FBI said in a FLASH alert late last week that a range of attackers have used Artex since at least August against organizations in many sectors around the world. However, the alert only summarizes 11 earlier reports and describes no new incidents.