# FBI WARNS AI HACKING TOOL IS BREAKING INTO NETWORKS IN ‘UNDER THREE HOURS’
**By Felix | November 17, 2026**
The FBI has put the security industry on notice: a cheap, open-source AI tool can now turn any large language model into a cyberweapon, and it is moving faster than defenders can react.
In a FLASH cybersecurity advisory released Friday, the bureau warned that hackers have used **Artex**, an open-source AI-powered penetration testing framework, to automate attacks across the U.S. and globally. Some intrusions advanced from initial targeting to **privileged access in under three hours**.
The advisory, dated November 13, 2026, links Artex to a cascade of incidents: a Chinese state-backed supply-chain attack that siphoned Microsoft 365 environments from more than 2,250 companies, and a wave of breaches at South Korean financial institutions that exposed the personal data of roughly 68,000 people.
how to vet an AI coding agent but for attackers. For developers and security teams, the message is blunt. The barrier to running autonomous cyberattacks just dropped to the price of an API key.
—
## What Is Artex, and Why Is It Different?
Artex is one of several open-source AI-powered penetration testing platforms designed to probe a network’s defenses. What sets it apart, according to the FBI, is its elasticity. Hackers can connect Artex to hosted frontier large language models such as those from Anthropic and OpenAI, or to open-source models and non-U.S. models.
The tool automates the full kill chain: reconnaissance, vulnerability discovery, exploitation, privilege escalation, and data theft, with limited human intervention.
The FBI’s advisory notes Artex has been used in cyberattacks since its first public debut in July 2026, with activity documented since at least August. It has targeted financial services, government, technology, defense, phone and internet companies, and law firms across multiple continents.
“You will often arrive after the attacker has finished.”
That warning, attributed to Barracuda’s chief information security officer, captures the central complaint of the advisory. Defenders who rely on spotting malicious activity and then responding will routinely arrive after the attacker has already moved on.
—
## South Korea: When a Single Attacker Hits Three Big Banks
The most visible campaigns involved South Korea’s financial sector. The FBI said a Chinese-speaking attacker combined Artex with trust in the AI era in mind — **Claude Code** agents to target Shinhan Bank, KB Kookmin Bank, and Hana Bank.
The attacks exploited weak cybersecurity defenses, including the absence of multi-factor authentication, and allowed hackers to steal the personal data of at least 68,000 people.
Separately, a South Korean cybercriminal gang used Artex in intrusions across more than 90 organizations. South Korean police arrested seven suspects linked to those attacks, which stole approximately 8.2 million data records spanning hospitals, educational institutions, and businesses.
The scale matters. This is not a lone script-kiddie guessing at default passwords. This is a coordinated campaign with state-level tooling deployed against civilian infrastructure.
—
## The ConnectWise Supply Chain: A Foot in the Door
The most consequential incident, however, was the supply-chain attack on ConnectWise, the IT management giant.
Microsoft attributed the October breach to a Chinese state-backed group it calls **Storm-0940**. Using Artex, the hackers compromised ConnectWise application identities, then attempted authentication against more than 2,250 downstream Microsoft 365 tenant environments.
In some cases, the intrusions advanced to credential changes and attempts to establish persistent access.
The attack exploited the cloud backup service formerly known as SkyKick Cloud Backup. The FBI’s advisory cites it as the October ConnectWise Cloud Backup supply-chain breach that allowed access to more than 2,250 downstream tenants.
Firewalls at the vendor were breached, not the destination. And that is the pattern the FBI keeps repeating.
—
## The Barracuda Lesson
Cybersecurity company Barracuda also disclosed that it was breached as part of an Artex-linked attack. In that case, the tool found an **exposed credential** belonging to a third-party service that Barracuda had not revoked after ending work with the vendor.
That access allowed hackers to reach stores of credentials Barracuda used to collect data from its customers’ systems.
Barracuda’s chief information security officer, Arve Kjoelen, wrote that “defenders who rely on spotting malicious activity and then responding to it will often arrive after the attacker has finished.”
Barracuda chose to “assume the worst” and changed all of its customer credentials.
—
## The Open-Source Problem
Artex’s developer responded to the damage by making the software closed-source. But the open-source version continues to circulate, and the FBI says it remains widely accessible for use in cyberattacks.
The exposure problem is compounding. Last week, researchers found hundreds of publicly exposed Artex systems online, placed on the internet without passwords, likely by mistake or misconfiguration. That included the Artex system used to target South Korea.
The tool has become a black market commodity: powerful, easy to wield, and trivially simple to leave unprotected.
—
## What This Means for Developers and Businesses
The Artex campaign arrives at a moment when AI agents are moving from experimental toys to daily productivity instruments. The same frameworks that help us write code, analyze data, and automate routine work are now being repurposed to exploit it.
Three lessons stand out.
**First, identity is the new perimeter.** Microsoft’s Storm-0940 campaign succeeded not by breaking encryption, but by stealing application identities. The ConnectWise breach began with compromised credentials, not a zero-day. Every service you use should have short-lived credentials, just-in-time access, and clean revocation policies.
**Second, the network is a single house, not a set of rooms.** Artex’s automated reconnaissance scans the whole property at machine speed. If one door is left unlocked, the attacker walks through, finds the keys, and moves on. Network segmentation, least-privilege access, and continuous monitoring are no longer optional decorations.
**Third, multi-factor authentication is a seatbelt, not a shield.** The South Korean banks lacked MFA, and that failure is directly cited in the advisory. It cost them data, reputation, and — in the case of the connected supply chain — a range of downstream customers.
The FBI’s guidance is practical: secure internet-facing applications, restrict cloud administration access, review permissions and authentication logs, and apply rate limits to sensitive APIs. Watch for bursts of automated requests, unexpected credential changes, and large data transfers.
—
## The Bigger Picture: The Agentic Arms Race
This is the first time the security industry has had to defend a category of attacker that can think.
Traditional malware follows a script. An AI agent follows a model: it can interpret, improvise, and chain techniques together at machine speed. The Artex framework has been used to automate multiple phases of cyber intrusions “from reconnaissance through data exfiltration,” according to the FBI’s flash advisory.
The result is a widening gap. Attackers need only one vulnerability, one exposed credential, one forgotten integration. Defenders need to get every single one right, all the time.
your agent will pick the hackable one is no longer measured in stolen credit cards. It is measured in the trust that companies place in the very AI tools they are adopting. When a developer points an LLM at a network and walks away, they are not just getting work done. They are raising the ceiling on what a determined attacker can do.
—
## Bottom Line
The Artex advisory is not a scare story. It is a mirror held up to the industry’s own habits. Identity hygiene, patch discipline, multi-factor authentication, and the exercise of least privilege are table stakes that too many organizations still treat as luxuries.
The FBI wants organizations to secure internet-facing applications, restrict cloud administration access, review permissions, and apply rate limits to sensitive APIs. Those are the same recommendations security professionals have been making for years. The only difference is urgency.
The AI graveyard is already open. The tools are faster now. The hours to compromise have shrunk. But the fundamentals have not changed, and nor should the fundamentals.
*Featured image: hacker laptop computer via Wikimedia Commons (CC0)*
Artex: Open-source AI-powered penetration testing framework used by the FBI in a flash advisory (Nov 13, 2026)
Storm-0940: Chinese state-backed group that used Artex in the October ConnectWise supply-chain attack
ConnectWise: IT management giant breached via supply chain; 2,250+ downstream Microsoft 365 tenants affected
South Korea: Three banks (Shinhan, KB Kookmin, Hana) hit with ~68,000 people’s data stolen; 7 suspects arrested, 8.2M records across 90+ organizations
Barracuda: Breached via an exposed third-party credential the company failed to revoke