Papalocal
Loading…
News
← All news
Technology

FBI warns AI hacking tool ARTEX can breach networks in under three hours

An open-source AI framework built for security testing has been turned against financial firms, government agencies, and tech companies worldwide — and removing it from GitHub hasn't stopped it.

How does this story make you feel?

Attackers have turned ARTEX — an AI-powered hacking framework originally designed for authorized security testing — against networks in the United States and around the world, the FBI said in a formal cybersecurity advisory. In some cases, the bureau warned, intruders moved from choosing a target to gaining privileged access in fewer than three hours.

The FBI's 13-page FLASH advisory, released on November 13, says multiple actors have weaponized ARTEX since at least August, hitting sectors including finance, government, technology, defense, and healthcare. Written in Go, ARTEX connects to external large language models — among them Claude, GPT, DeepSeek, GLM, Grok, and Kimi — and splits attack objectives between a planning agent and worker agents that execute tasks with little human involvement, the bureau said.

The advisory corroborated earlier findings from Microsoft: a China-state-aligned group leveraged ARTEX to penetrate IT management company ConnectWise, then reached the Microsoft 365 cloud environments of at least 2,250 ConnectWise customers and took data from those accounts. TechCrunch reported that a ConnectWise spokesperson did not respond to a request for comment.

ARTEX was also used to hack a number of South Korean financial institutions, with the FBI attributing the breaches in large part to weak cybersecurity defenses and absent protections such as multi-factor authentication. The personal data of at least 68,000 people was exposed as a result of those attacks.

Cybersecurity company Barracuda separately disclosed it was breached in an ARTEX-linked attack by financially motivated criminals. The tool found an exposed credential belonging to a third-party vendor whose access Barracuda had not revoked after ending the business relationship. That foothold gave attackers access to credentials Barracuda used to collect data from its own customers' systems. Barracuda said its logs were not detailed enough to rule out data access and chose to assume the worst, changing all customer credentials. According to TechCrunch, a Barracuda spokesperson said the company directly notified and advised a minority of its XDR customers to revoke and recreate integration credentials, without specifying how many.

Defenders who rely on spotting malicious activity and then responding to it will often arrive after the attacker has finished.— Arve Kjoelen, Chief Information Security Officer, Barracuda

The speed of ARTEX attacks is central to why the FBI considers them a serious escalation. The tool typically begins by rapidly scanning a victim's network for weaknesses, then exploits simple security flaws — weak passwords, exposed credentials, default admin settings — to gain a foothold. From there, it spreads across the network hunting for databases, mailboxes, and other sensitive information. While the attacks can generate large volumes of suspicious traffic that may trigger defensive alerts, they move faster than human security teams can typically investigate and act.

Amazon's threat intelligence team published its own findings three days after the FBI advisory, adding a dimension the bureau's report did not fully detail: ARTEX agents have been actively hunting for exposed AI applications and stealing the API keys stored inside them. Amazon captured records of this activity using MadPot, its network of decoy servers posing as popular AI services. The agents extracted AI provider keys, verified they worked, and pooled them — using stolen compute credits to fund further attacks. Amazon said MadPot began capturing this agent traffic in early October.

Drawing on Simplified Chinese prompts and China time-zone settings found in captured records, Amazon said it was moderately confident the agents operated on ARTEX and that the operators were distinct China-linked groups. Regarding the outcomes the agents themselves reported, Amazon said independent verification was not possible. The captured records named AI platforms including One API, New API, Dify, FastGPT, Langflow, and Ollama as targets, with hundreds of targets appearing in the data.

Cloudflare also reported ARTEX activity involving a compromised account-level CI/CD API token that carried no IP restrictions or rate limits. Within one hour, the activity included 400 database dump requests. Cloudflare identified static, non-expiring tokens in build pipelines as the common root cause, according to Daily Security Review.

The original ARTEX repository, published on GitHub in July by a developer using the handle 'Autumn-27,' was removed on October 8 — apparently after the tool began appearing in real attacks. The FBI has said, however, that the earlier open-source version keeps circulating and is still widely within reach. Researchers counted more than 1,600 exposed ARTEX consoles online as of early November, some left without password protection by misconfiguration — including the console used in the South Korean attacks. According to Daily Security Review, forks, archived copies, localized versions, and a commercial variant called 'ARTEX PRO' are also still obtainable.

On the same day the FBI released its advisory, the malware researcher group MalwareMustDie published an open-source detection signature pack. Amazon notified potentially affected AWS customers. The FBI's advisory included indicators of compromise for defenders.

The pattern illustrates a compounding risk: AI infrastructure holds credentials that are directly monetizable, and in this campaign those stolen credentials also convert into more attack capacity, creating a self-funding loop. Amazon pointed to the agents' low skill barrier and speed as the central concern, adding that the removal of the original repository has done little to limit distribution once forks exist.

Why it matters — ARTEX lowers the skill and time required to breach organizations to the point where stolen AI credentials automatically fund more attacks, and removing the tool from GitHub has not stopped its spread — meaning the threat will persist regardless of takedown efforts.

⚠ Not yet confirmed

  • Amazon assessed with moderate confidence that ARTEX agents were operated by separate China-linked groups, based on Simplified Chinese prompts and China time-zone settings
  • Cloudflare's ARTEX-linked database dump requests resulted in data being exfiltrated

Reported by techcrunch.com, malwarebytes.com, techspot.com, dailysecurityreview.com

Community

Be the first to comment.

Share this story
Text Email