A new report from Amazon Threat Intelligence has described how it recently detected malicious campaigns that it suspected were run autonomously by the ARTEX AI penetration testing platform. According to the report, this indicated that agentic workflows “have progressed from experimental use to sustained, multi-objective offensive operations that run with minimal human intervention.”
Amazon detected the cyber operation thanks to its MadPot network of decoy sensors that were disguised as popular LLM inference APIs. These received requests from AI agents that included the agents’ own campaign records, giving the defenders a window into the attackers’ plans.
After analyzing the agents’ actions, Amazon concluded that the most prominent use of ARTEX was a large-scale theft of AI model credentials. Any stolen keys that were discovered were used to fund further credential hunting, making the operation somewhat self-sustaining.
Another ARTEX campaign sought to bypass know-your-customer (KYC) verification at cryptocurrency exchanges by using “real-time deepfake injection.”
Amazon’s report noted that it attributed the campaign to ARTEX with “moderate confidence” based on “prompt references to ARTEX tooling and other technical artifacts referencing ARTEX.” However, it leaves open the possibility that some other technique was used instead.
As for who was responsible, Amazon said it had moderate confidence that the hackers were “disparate Chinese-nexus threat groups” due to the presence of Chinese-language prompts in the operation, as well as Chinese time-zone configurations, toolsets popular with Chinese threat actors, and “targeting consistent with Chinese interests.” That included certain “red lines” in the ARTEX instruction sets, such as excluding targets in mainland China and government domains.
Amazon’s report advised defenders that while the observed techniques were not new, the speed and persistence of the AI agents changed the equation. “The agents succeeded where basic security hygiene was missing, including default or hardcoded credentials, keys left in code and public artifacts, unauthenticated interfaces and inference endpoints, and unpatched software,” Amazon warned. Fixing those weaknesses is key to keeping attackers out.
And considering how swiftly AI agents can move, Amazon noted that “organizations should assume exposed secrets and misconfigurations will be found quickly and remediate accordingly.” Any AI applications, gateways and endpoints should be controlled in the same way as other production systems, while organizations must “watch for misuse of AI services through unexpected spend or unusual usage.”
The campaign came just as the FBI warned that ARTEX is powering cyberattacks around the globe, and that it’s not just limited to one particular threat group. It suggests that this kind of cyberattack could become increasingly common in the future, making it more important than ever for defenders to keep their systems well protected.