Three weeks ago we wrote about ARTEX, an open-source AI-based penetration testing framework, after WorkOS identified it in an attack on an internal system. At the time it sounded like a strange story about an experiment gone wrong. Since then ARTEX has become one of the most talked-about attack tools in the world: it appears in an official FBI alert, in the breach of security company Barracuda, and according to one report also in an attack on the European Commission.

What does the FBI say?

In a FLASH alert issued on November 13, the FBI describes ARTEX being used since at least August against organizations in finance, government, defense, technology, healthcare, energy and telecoms. The most disturbing figure is speed: according to the alert, in some cases attackers went from targeting to administrator privileges on a network in under three hours. In South Korea, according to reports, several financial institutions were breached, mainly because they hadn't enabled multi-factor authentication, and personal data of about 67,000 people was stolen.

How was Barracuda breached?

Barracuda's story shows how simple the weak point can be. According to the company, it once gave a cloud account access key to an AI vendor for a project and never revoked it. The attacker found the key in a configuration file in the vendor's GitHub repository and, over about 44 hours on November 1 and 2, moved across about ten cloud accounts, reached secret stores and left a backdoor. According to Barracuda, the activity patterns match ARTEX, but it has not attributed the attack to a specific group, because the tool is available to anyone.

And in Brussels too

On November 19 the European Commission confirmed that an attack detected in October was carried out with a publicly available AI tool. According to reports, several of the Commission's internal systems were disrupted for about 10 days, and the Commission said it found no evidence of a breach of data on staff as a whole. The Commission did not name the tool. According to Politico, CERT-EU, the cyber team of the EU institutions, identified ARTEX in it, but this was not officially confirmed. It is, according to reports, the first time the EU has publicly confirmed an attack carried out with AI.

Why is this different?

Penetration testing tools have existed for years, and attackers have always used them. The difference is autonomy: ARTEX doesn't just run scans a person chose, it decides on its own what to check, when to dig deeper and how to move through a network. That means an attacker with little knowledge can achieve a professional team's results, and a professional attacker can run dozens of attacks in parallel. According to reports, researchers found more than 1,600 publicly exposed ARTEX management consoles online, and the tool's developer removed the repository after the attacks. But open-source code that has already spread doesn't really disappear.

What do the cases have in common?

In almost every one, the tool didn't break through an impenetrable wall. It found an open door: institutions without multi-factor authentication, a cloud key forgotten in a code repository, permissions never revoked. AI doesn't invent new weaknesses, it finds the old ones faster than any person. That is exactly what makes it dangerous: organizations that relied on "no one will find that" discovered a machine does.

What should organizations do now?

First, the basics: multi-factor authentication for all remote access and every privileged account. Second, key management: map every access key given to vendors, revoke unused ones, limit the permissions of the rest, and scan code repositories, including vendors', for exposed secrets. Third, response time: if an attacker reaches admin privileges within three hours, monitoring that checks alerts once a day is no longer enough.

And the Israeli angle

For Israel's cyber industry this is both a threat and an opportunity. Israeli organizations, like any organization, are exposed to the same tools. On the other hand, this wave explains the money flowing in recent months to AI-based defense companies: Corma raised at a $1 billion valuation to build a defensive AI model, and Bay and Daylight raised tens of millions. When attackers use agents, defenders need agents too.

The bottom line

ARTEX isn't the last tool of its kind, it is the first to make headlines. The takeaway shouldn't be panic but a return to basics: closing open doors before a machine finds them. Organizations that get the basics right will be a far less rewarding target, even for an attacker that never sleeps.