Skip to main content
Emerging ThreatsMalware & Ransomware

FBI and AWS Warn of AI-Powered Cybercrime Surge

Dimly lit server room with rows of computer equipment and softly glowing indicator lights, a single out-of-focus laptop…

"Agents validated identities through each provider’s token and identity services, enumerated storage, assessed permission scope, and went after managed AI model services, secrets and parameter stores, and application identities," AWS Chief Information Security Officer CJ Moses wrote.

How ARTEX automates intrusion at "industrial scale"

Security alerts from the FBI and Amazon Threat Intelligence describe ARTEX, an open-source AI penetration testing framework, being repurposed by attackers to automate multi-stage intrusions. Between October 3 and November 5, Amazon’s MadPot honeypot sensors detected two distinct ARTEX campaigns that illustrate automation across credential theft, model access, and identity-verification bypasses. The alerts stress that while the techniques themselves are not new, ARTEX lets attackers chain them faster and more persistently.

Two campaigns detected by Amazon MadPot honeypots

Amazon’s sensors recorded one campaign that used AI agents to steal API keys “at industrial scale” and a second that focused on defeating digital identity checks. In the credential-theft campaign, miscreants harvested API keys and fed them into ARTEX’s model gateway to power more agents searching for further access. In the deepfake campaign, attackers used ARTEX to generate real-time deepfake videos to try to trick live selfie (liveness) checks at two cryptocurrency exchanges and to forge ID photos for document checks.

Targets named: LLM gateways, proxy platforms and AI builders

The alerts list specific targets. ARTEX agents went after LLM gateways and proxy platforms including One API, New API, Veloera, Sub2API, and LiteLLM, and targeted AI application builders Dify, FastGPT, Langflow, RAGFlow, Open WebUI, LobeChat, and Ollama. AWS reported the campaign also used cloud credentials to gain stolen access to models and that agents pursued "managed AI model services, secrets and parameter stores, and application identities." The FBI and AWS note the attackers succeeded where basic security hygiene was missing — default or hardcoded credentials, keys left in code and public artifacts, unauthenticated interfaces and inference endpoints, and unpatched software.

Deepfakes, liveness protocols and a persistence test

AWS detailed how agents reverse-engineered a vendor’s liveness protocol "in detail, documenting the face-detection event stream it sends, how it tracks face position during the check, and the coordinate system it uses to tell the user where to move." The agents reproduced the full movement sequence with the face positioned within 7 pixels of the target, but the vendor's server-side verdict rejected the deepfake every time. In another session, attackers built a real-time deepfake video pipeline on a single consumer-grade GPU using an open-source face animation model that ran at 11.5 frames per second — "close to what a live video check expects." Agents also fabricated camera metadata whose image dimensions matched the JPEG file and preserved original camera compression tables that usually change when an image is re-saved.

What technologists and security teams, policymakers, and affected enterprises should do

  • Technologists and security teams: The FBI and AWS emphasize fundamentals — remove default/hardcoded credentials, eliminate keys in code or public artifacts, authenticate interfaces and inference endpoints, and keep software patched. The FBI advised restricting sensitive services to trusted networks, private connectivity, or dedicated administrative paths and moving enterprise applications behind zero-trust access policies and permissions.
  • Policymakers and regulators: The FBI's Flash alert formalizes a list of mitigations for internet-exposed applications, self-service portals, cloud infrastructure, and partner or supplier systems. Regulators reviewing critical-sector resilience should note that automation reduced some intrusions "from initial targeting to privileged access in under three hours," according to the FBI.
  • Affected enterprises and procurement leaders: The alerts name concrete supply-chain and downstream impacts: an August incident in South Korea stole personal data for about 67,000 people, and a September ARTEX-linked campaign that targeted ConnectWise Cloud Backup customers ultimately compromised more than 2,250 downstream Microsoft tenants. Procurement and risk teams should evaluate vendor exposure and consider access controls to model gateways and AI application builders listed in the alerts.

The FBI noted that after public breaches, the original ARTEX developer removed the GitHub repository in October, but forked and localized variations plus a commercial version remain. As of November 4, internet scanning service Censys identified 1,465 hosts running an exposed ARTEX console — a figure the FBI said is "likely significantly higher when including systems behind firewalls or in private networks." The agency urged organizations to implement its Flash alert mitigations promptly.

The technical details in the AWS and FBI notices point to a clear operational lesson: automation via AI agents amplifies old weaknesses. Where basic hygiene fails, ARTEX-style agents can escalate and diversify attacks in hours rather than days. Whether defenders can close those gaps at enough organizations — and how quickly internet-exposed AI gateways and application builders are hardened — will determine whether these campaigns remain isolated incidents or become a new routine for opportunistic attackers.

https://www.theregister.com/cyber-crime/2026/11/16/attackers-are-using-ai-agents-to-automate-cybercrime-at-industrial-scale-fbi-and-aws-warn/5310223