Vulnerability  ·  2026-11-15

FBI FLASH-20261113-001: ARTEX autonomous AI penetration-testing framework weaponized in the wild (agentic AI intrusions + AI-model-credential theft)

VulnerabilityHigh impactGlobal
On 2026-11-13 the FBI issued FLASH-20261113-001 (and an IC3 Threat Overview PDF) warning that ARTEX, an autonomous AI penetration-testing framework, is being used by state-backed and criminal actors to automate multi-phase intrusions 'at machine speed.' The advisory documents operations observed since August 2026 spanning recon, exploitation, privilege escalation and exfiltration, and notes attacker-controlled deployments favor Chinese models (predominantly DeepSeek Flash) for capability, low cost and limited observable guardrails when framed for pentesting.
This is a government-confirmed, in-wild active campaign built on agentic AI — an AI agent framework autonomously driving real intrusions. For AI defenders specifically, the same framework is being used to steal AI-provider API keys and model-account credentials, so the attack surface now includes an attacker turning the victim's own AI-budget credentials against them (LLMjacking), not just classic infrastructure.
Threat actors host their own ARTEX consoles and drive autonomous agents that chain reconnaissance, exploitation (unsigned SAML assertions, XXE, SSRF, SSTI, SQLi, command injection, anonymous uploads), privilege escalation and exfiltration at machine speed; some intrusions go from initial targeting to privileged access in under 3 hours. The FBI and independent research note attackers use it specifically to harvest AI-model API credentials (LLMjacking), using the stolen compute to fund further inference.
ARTEX autonomous penetration-testing framework (developer 'Autumn-27'; GitHub repo pulled Oct 8 but forks/mirrors/ARTEX PRO persist); also the AntiFox fork. Connects to Claude, GPT, DeepSeek, GLM, Grok, Kimi as the reasoning backend.
Inventory every internet-facing app; keep cloud admin/backup/SCM/Kudu endpoints off the open internet; enforce least privilege on multi-tenant app consent grants; rate-limit high-impact business APIs; scan code/pipelines for embedded secrets; review every Microsoft 365 tenant where ConnectWise backup apps were ever authorized; investigate unexpected LLM API traffic. See FBI FLASH-20261113-001 / IC3 Threat Overview (https://www.ic3.gov/CSA/2026/261113.pdf).
IC3 / FBI Threat Overview of ARTEX Autonomous Penetration Testing (FLASH-20261113-001)GBHackers — FBI Warns Hackers Are Weaponizing ARTEX AI FrameworkMLQ.ai — FBI warns ARTEX AI hacking tool can reach privileged access in under three hoursOODA Loop analysis of the FBI FLASH on ARTEX
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →