What we know so far: The FBI says that hackers are using an AI-driven penetration testing tool called Artex to automate attacks that used to require skilled human operators. The software connects to large language models and handles the scanning, exploitation and lateral movement on its own, and in some cases it has gone from first targeting a system to active compromise "in under three hours."

Artex belongs to a growing category of open-source AI penetration testing platforms. Victims linked to the tool so far include IT management firm ConnectWise and thousands of its customers, security vendor Barracuda, and a group of South Korean financial institutions.

Security teams use this kind of software to probe their own defenses before attackers do the same. The FBI says attackers have taken the same capability and turned it outward against financial services firms, government agencies, technology and defense companies, telecom and internet providers, and law firms.

The tool's design explains much of its appeal. Operators can connect Artex to hosted frontier models from Anthropic and OpenAI, open-source models, or models built outside the US. Once connected, Artex usually begins by scanning a target network quickly for weak spots. It then goes after simple flaws, such as poor protections and exposed or stolen passwords, to get an initial foothold. From there it moves toward databases, mailboxes and other stores of sensitive information.

The FBI describes Artex intrusions as "noisy," meaning they produce large volumes of traffic and data that can set off a victim's defensive alerts. Speed makes up for the noise, though. The tool works faster than most human responders can investigate, so an alert may arrive while the attacker is already spreading through the network.

Artex was first released publicly in July, and the advisory ties it to several campaigns since then. In South Korea, attackers used it against a number of financial institutions, and the breaches were blamed largely on weak security, including missing multi-factor authentication. Personal data belonging to at least 68,000 people was stolen.

The biggest known operation involved ConnectWise. The FBI confirmed Microsoft's earlier finding that hackers used Artex to breach the company and reach the Microsoft 365 cloud environments of at least 2,250 of its customers.

Microsoft has said the attackers are likely Chinese, and they used the tool to steal data from the affected customers. ConnectWise did not respond to a TechCrunch request for comment on the FBI's findings.

Barracuda's breach shows how small the opening can be. The company said financially motivated criminals using Artex found an exposed credential belonging to a third-party service that Barracuda never revoked after it stopped working with that vendor.

With that credential, the attackers reached stores of credentials Barracuda uses to collect data from its customers' systems. The company said its logs were not detailed enough to rule out access, so it decided to "assume the worst" and rotated all of its customer credentials.

Arve Kjoelen, Barracuda's chief information security officer, described what this means for traditional security operations. "Defenders who rely on spotting malicious activity and then responding to it will often arrive after the attacker has finished," he wrote.

A Barracuda spokesperson told TechCrunch that the company "directly notified and advised a minority of XDR customers to revoke and recreate integration credentials." The spokesperson did not say how many customers were affected.

After the attacks, Artex's developer made the software closed source to stop further misuse. The FBI says the change came too late to prevent further misuse. The earlier open-source version is still circulating and remains "widely accessible," so anyone with a copy and access to a capable model can still run it.

The tool's own infrastructure has also proved poorly secured. Last week, researchers found hundreds of Artex systems exposed on the public internet with no password, most likely because of mistakes or misconfiguration. One of them was the system used in the South Korean attacks. Exposed instances like these could let outsiders take over ready-made hacking setups without building their own.

None of the intrusions described so far relied on new exploits. Artex succeeded by finding the same problems attackers have always used, such as missing multi-factor authentication, leaked passwords and credentials left active after a vendor relationship ended. What has changed is that an AI agent can now find and use those weaknesses faster than many defenders can react.