Attackers Automate Cybercrime with AI Agents, FBI and AWS Warn
Threat actors are leveraging the open-source ARTEX AI framework to automate cybercrime at an 'industrial scale,' stealing API keys and generating deepfakes, according to the FBI and AWS.

Attackers are increasingly weaponizing the open-source AI penetration testing framework ARTEX to automate cybercrime across critical sectors, including finance, government, and healthcare. Security alerts from the FBI and Amazon Web Services (AWS) detail two distinct campaigns observed between October and November, highlighting how AI agents are accelerating intrusions and enabling sophisticated attacks.
The first campaign focused on the large-scale theft of API keys, which attackers then used to gain unauthorized access to AI models. These stolen keys allowed the AI agents to enumerate storage, assess permissions, and target managed AI model services, secrets, and application identities. AWS noted that targets included popular LLM gateways and proxy platforms such as One API, New API, Veloera, Sub2API, and LiteLLM, as well as AI application builders like Dify, FastGPT, and Langflow. The attackers also exploited cloud credentials to facilitate access to these valuable AI resources.
A second observed campaign demonstrated the use of ARTEX to generate real-time deepfake videos and forged ID photos for identity verification bypass. Attackers reverse-engineered a vendor's liveness detection protocol, documenting its event stream and face-tracking mechanisms. While their deepfake videos, generated at 11.5 frames per second, were sophisticated enough to fool some visual checks, they were ultimately rejected by the vendor's server-side verdict. The agents also focused on document checks, creating synthetic camera metadata to mimic legitimate images.
These campaigns underscore how AI agents can significantly boost the speed and persistence of cyberattacks. The FBI highlighted that ARTEX can reduce the time from initial targeting to privileged access to under three hours. This acceleration was evident in past incidents, such as August attacks on South Korean financial institutions that compromised personal data for approximately 67,000 individuals, and a September supply chain attack targeting ConnectWise Cloud Backup customers, which ultimately affected over 2,250 downstream Microsoft tenants.
While the underlying attack techniques are not new, the integration of AI agents amplifies their effectiveness. The FBI pointed out that these attacks succeeded due to fundamental security lapses, including exposed credentials, keys left in code, unauthenticated interfaces, and unpatched software. "Getting those fundamentals right remains the most effective defense," AWS Chief Information Security Officer CJ Moses stated.
The ARTEX framework's developer removed the original GitHub repository in October following these disclosures. However, forked versions and commercial offerings remain available. Internet scanning services identified hundreds of hosts running exposed ARTEX consoles, with the FBI estimating the total number of deployments to be significantly higher, especially within private networks.
In response, the FBI urged organizations, particularly those with internet-exposed applications, cloud infrastructure, and supplier systems, to implement robust mitigation measures. These include reducing external attack surfaces, restricting sensitive services to trusted networks, and enforcing zero-trust access policies. The coordinated warnings from the FBI and AWS signal a growing concern over the industrialization of cybercrime through AI automation.
The implications extend beyond direct attacks, as the ability to automate complex tasks like deepfake generation and credential theft at scale poses a significant challenge to existing security paradigms. The rapid evolution of AI-powered tools necessitates a proactive and adaptive approach to cybersecurity, focusing on fundamental security hygiene and advanced threat detection capabilities.
The FBI's warning specifically highlights that ARTEX has been used by attackers since at least August, targeting sectors including finance, government, healthcare, and technology. Additionally, Microsoft reported a China-state-aligned group leveraging ARTEX to accelerate attacks after compromising ConnectWise application credentials, providing further attribution and scope to the threat.