Artex is a free tool from China that uses a team of AI agents to probe a company's systems for weaknesses, the way a hired security tester would. Criminals and state-linked hackers have used it in real attacks, and hundreds of copies of it are open to anyone on the web with no password. Its developer announced in October that there would be no more public releases, and that changes less than it sounds like.
Closing the code does not call the copies back
Anyone who already downloaded Artex keeps it, and copies of its code are still on code-sharing sites. The tool needs no license, no payment and no account with its maker, so nobody can switch it off. Pirated Cobalt Strike was in the same position, and the clean-up there worked by finding the servers the copies ran on: in 2023 law enforcement took action against hundreds of installations, and illegal use fell over the following two years.
Artex has the same weak point, and its users are making it easier. More than 30 of the exposed copies have open folders with attack tools inside them. The hacker behind the September attacks on Korean financial companies was found because of exactly that mistake, and that server's folders had been open for seven weeks before the attack was reported. The tool that makes attackers fast also makes some of them careless, and defenders get a view of them that they rarely have.
What the agents actually try
Artex does not invent new ways in. It runs the old ones without getting tired. Its agents scout what a company has online, guess passwords across many accounts, check what a captured account can reach, and copy out what they find. It can use any AI model for the thinking, so the attacker brings the model and Artex supplies the routine.
In the October attack on ConnectWise customers, the attackers began with a stolen vendor key. Artex then tried to sign in to more than 2,250 companies' Microsoft 365 systems, at times 2,900 tries a minute, and some attacks reached high-level access in under three hours. A person at a keyboard manages a few tries a minute. Most of the targeting still failed, because a tool that tries everything at machine speed has to find something open.
What it means for a firm of forty people
A small firm used to be safe by being small. A human attacker has to spend hours on each target, so they pick the rich ones. An agent spends almost nothing on one more target, so a 40-person accounting firm lands on the same list as a bank, which is what attackers trying every side door already looked like. The agent tries a few common passwords on every mailbox it can find. If the shared finance mailbox has a password made of a season and a year, and no code is sent to a phone, that is the account it gets into.
Because the copies will last for years, the cost of probing a business will stay close to zero. The firms that get breached will mostly be the ones where the basics were off.
Start with the accounts that can do the most damage: the administrator, the finance mailbox and any login shared by several people. Give each one a long, unique password and a second sign-in step, such as a code or an approval on a phone. These are the first places an agent like this goes, and for a small firm that is an afternoon's work.