Beacon: Hundreds of Artex AI Hacking Tools Found Exposed Online

This story appeared in our weekly cybersecurity newsletter Beacon. Sign up to get weekly updates on the latest news, findings and insights, straight to your inbox every Thursday at 10AM.

Top Story This Week:

Hundreds of Artex AI Hacking Tools Found Exposed Online

Researchers have found more than 800 deployments of Artex, an open-source autonomous AI penetration testing tool, accessible from the internet with no password needed. Not every instance is malicious, and many belong to authorized testers, but it is unclear why so many are exposed. The exposures also reveal attack tooling such as implants. Released in July and free to download, Artex is a popular option for people who can’t access the most advanced AI security models. Microsoft describes it as a system of AI agents that can run an attack from reconnaissance through to data theft with little human input.

The tool has been linked to a growing list of campaigns, including attacks on South Korea’s financial sector and the ConnectWise compromise that exposed data held in Microsoft’s cloud by thousands of organizations. Microsoft says the attackers stole data so quickly that defenders had little time to detect and respond, and reports stated more than 200 organizations targeted globally since mid-October. Many of the attacks exploited simple weaknesses, such as weak or known passwords and missing multi-factor authentication.

Separately, Microsoft Threat Intelligence says attackers have been using Artex and other agentic AI frameworks since August to scan public GitHub repositories for exposed credentials and API keys. From October 28, one IP address used still-valid application credentials to sign in to downstream tenants, stealing data from Azure Key Vaults and Storage Accounts. Microsoft’s advice is to assume any secret exposed in a public repository is compromised. Artex has announced it is going closed-source, but dozens of copies remain on code-sharing sites.

Why it Matters

Artex shows what happens when attack tooling gets AI agents. Reconnaissance, exploitation and data theft run in sequence at machine speed, and one free tool can be pointed at hundreds of organizations at once.

Defenders working through alerts, scans and patch cycles at human pace are being outrun, as Microsoft’s account of the ConnectWise attacks shows. The speed is the new part, but the entry points often aren’t: weak passwords, missing MFA, misconfigurations and credentials left in public code repositories.

An AI agent doesn’t need a clever exploit when the basics are open. Organizations need to see their attack surface as automated attack tooling does, including what staff and developers leave in public places, and fix the exposures that matter most first.

Advice

The speed and scale at which AI-enabled tools can operate mean that relying on detection and response alone is becoming increasingly risky. By the time an automated attack is identified, attackers may already have moved through multiple stages of an operation. Organizations need to get ahead of this activity with a preemptive approach, identifying and removing the opportunities attackers are likely to exploit before they find them.

That means continuously understanding your external attack surface, including shadow IT and third-party connections, and validating which exposures are genuinely reachable and exploitable. It also means looking beyond vulnerabilities to the behaviors and techniques attackers use, assessing attacker TTPs drawing on threat intelligence to anticipate how tools like Artex might be deployed against your environment.

Crucially, this doesn't mean overlooking the fundamentals. Strong authentication, MFA, least privilege and secure configurations remain essential, particularly when relatively simple weaknesses can provide automated tools with an easy route in. The objective should be to reduce the opportunities available to attackers, rather than relying on your ability to respond quickly enough once they've found them.

Discover More

AI vs. the Patch Window: What has Changed, And What Hasn’t

Our latest report cuts through the hype to examine how AI is accelerating vulnerability discovery, exploitation and disclosure. The biggest shift isn’t a doomsday scenario or a brand-new class of attack. It’s an acceleration of trends teams were already struggling to keep up with. Waiting for disclosure, then patching within a comfortable window, is getting harder to sustain. Read more.

The Complete Guide to Preemptive Cybersecurity

Preemptive cybersecurity is a security strategy focused on identifying and closing exposures before attackers can exploit them, rather than detecting and responding to attacks after they’ve already begun. As exploitation timelines collapse from weeks to hours, the shift from reactive defense to proactive exposure reduction has become the defining challenge for modern security teams. Read more.

Weekly News Digest

Hackers Pull off $133 million French Bank Heist via Service Provider

French banking group Credit Mutuel Arkea confirmed that threat actors exploited access through an unidentified service provider’s platform to initiate fraudulent transactions using the bank;s own account. The bank estimated losses at approximately $133 million. Some funds were converted into cryptocurrency. This incident demonstrates how compromise of a trusted third party platform can bypass a financial institution’s primary security controls and enable direct thrift. The movement of funds into cryptocurrency will complicate recovery.

3AM Hits European Schools: A Hybrid AI Ransomware Attack, Part Human, Part Agentic Loop

Bitdefender describes the attack by a 3AM ransomware affiliate on a UK school estate as a hybrid operation; part-human operator, part-AI agent driven. This assessment is primarily based on aspects of attacker behaviour that resembled an agentic loop, whereby a program feeds results back into an LLM model and runs the subsequent output. Analysis of the intrusion showed some commands being run and retried faster than would be feasible for a person to do, while other actions had a longer gap between them implying human decision making. Bitdefender claims this hybrid operation is the first of its kind that they have investigated.

Asian Airlines Announce Data Breaches After Attack on Spanish IT provider Amadeus

Several major Asian airlines, including Japan Airlines, Cathay Pacific, Korean Air, Singapore Airlines and Malaysia Airlines, have reported data breaches following a cyber attack on Spanish IT provider Amadeus. The incident exposed tens of thousands of customer records, potentially including passenger names, booking references, flight details, and passport information. Amadeus confirmed that a vulnerability in its booking-display environment had been exploited but stated that there was no evidence of financial data being accessed. Although airline operations remain unaffected, customers have been warned to remain vigilant against potential phishing attempts using their exposed personal information.