AI·News & analysis
FBI warns hackers are using the ARTEX AI tool to attack organizations worldwide
A new FBI alert says criminals and state-linked groups use the open-source ARTEX framework to run AI agents that break into networks in hours, sometimes with little skill needed.

Tide
Wave
Sci-fi
6/10
Reality
In your hands
Hacking crews that never need a coffee break.How we rate
The FBI warned that several threat actors have used the open-source ARTEX AI framework to automate cyberattacks since at least August 2026.
Some break-ins reached privileged access in under three hours. The tool needs little skill to run, and researchers found more than 1,600 exposed consoles. The developer deleted the code in October, but copies and a paid edition remain, so the FBI is urging defenders to act now.
What to know
- The FBI says multiple threat actors have used the ARTEX framework since at least August 2026, across finance, government, tech, defense and healthcare.
- Some intrusions went from first target to privileged access in under three hours.
- ARTEX does not have its own AI model. It plugs into outside models such as Claude, GPT, DeepSeek, GLM, Grok and Kimi.
- The developer deleted the code on October 8, but copies, forks and a paid ARTEX PRO edition are still around.
The FBI has warned that hackers are using an AI tool called ARTEX to run cyberattacks on organizations around the world. The twist is that ARTEX was built as a security tool.
So how does a tool made for defenders end up in the hands of attackers? Let's start with what it is.
First, what is ARTEX?
ARTEX is an autonomous penetration testing framework. Penetration testing means breaking into a computer system on purpose, with permission, to find weak spots before criminals do. Think of a locksmith you hire to try every window of your house.
ARTEX is written in Go, a popular programming language. It uses AI agents, which are small programs that make decisions and take steps on their own. They scout a target, look for flaws, exploit them, climb to higher access and steal data, with little human help.
A developer known as Autumn-27 released it on GitHub in July 2026. It is open source, which means anyone can download it, read it and change it.
How does it actually work?
ARTEX does not have an AI brain of its own. It plugs into outside large language models, or LLMs, the kind of AI that powers chatbots. The list includes Anthropic Claude, OpenAI GPT, DeepSeek, GLM, Grok and Kimi.
The operator types in goals. ARTEX turns them into commands and hands them to specialized AI agents. Each agent handles one job, and many can run at the same time.
That is how one person can go after several organizations at once. The FBI put it plainly: the framework "enables threat actors with minimal technical expertise to conduct complex, multi-target operations simultaneously."
In real life Imagine one burglar who can send ten helpers to ten houses at the same time. Each helper tries the doors, the windows and the garage, and reports back only when it finds a way in. ARTEX does that for computer networks.
What did the FBI say?
The agency released a 13-page FLASH advisory on Friday. A FLASH is a quick alert the FBI sends to defenders when it sees an active threat.
Here are the main points:
- Who: multiple threat actors, active since at least August 2026.
- Targets: financial, government, technology, defense, healthcare and other sectors.
- Speed: "some intrusions progressing from initial targeting to privileged access in under three hours."
- Weak spots: overlooked systems such as partner portals, employee apps, legacy services and third-party integrations.
The catch: these attacks make a lot of noise and can trigger security alerts. But they move so fast that defenders may have little time to react before the attackers are inside.
Where has ARTEX been used?
The FBI advisory points to several earlier cases. Together they show how far the tool has spread.
- ConnectWise, October: attackers compromised the application identities of ConnectWise's cloud backup service, formerly called SkyKick Cloud Backup. An application identity is like a badge that lets a program sign in to other systems. With those badges, the attackers could log in to customer Microsoft 365 accounts. This kind of break-in is called a supply-chain attack, because it goes through a supplier.
- South Korean banks: a Chinese-speaking threat actor combined ARTEX with Claude Code agents to attack banks including Shinhan Bank, KB Kookmin Bank and Hana Bank. The personal information of about 67,000 people was exposed.
- Police arrests: South Korean police arrested seven suspects linked to attacks on more than 90 organizations, and ARTEX was reportedly used in most of them. About 8.2 million data records were stolen from hospitals, schools and businesses.
- Vapor Panda: CrowdStrike reported that a Chinese cyberespionage group it tracks under this name used ARTEX against more than 200 organizations in 14 countries.
We covered the banking case earlier in our story on the South Korean bank hacks.
How did the ConnectWise attack unfold?
Microsoft traced the ConnectWise campaign to a China-state-aligned group it calls Storm-0940. The group used ARTEX to speed up reconnaissance, which means gathering information about a target, plus password spraying and abuse of cloud app permissions.
Password spraying is a simple trick. Instead of guessing many passwords for one account, the attacker tries a few common passwords on many accounts, so no single account locks up.
After compromising ConnectWise's identities, the attackers used ARTEX to try signing in to more than 2,250 downstream Microsoft 365 tenants. A tenant is one organization's private slice of a cloud service. Some intrusions went on to change credentials and try to set up persistent access, which means a way back in later.
Security firm Huntress had reported at least 162 compromised tenants at the time.
What did CrowdStrike find about Vapor Panda?
CrowdStrike's report gives the clearest look at what these AI agents do. The operation began on October 17. Targets included government ministries, telecom providers, and pharmaceutical, semiconductor, defense, nuclear and financial organizations.
The group compromised at least nine organizations. It also ran password spraying and credential scanning against at least 12 other high-value targets.
Confirmed victims include telecom providers in the United States and Belgium, government bodies in Greece, the Middle East and Latin America, and pharmaceutical firms in the US.
Some details stand out:
- Forged access: in the Greek case, the agents forged an administrator token and read mailboxes. A token is a digital pass that proves who you are. They built it from authentication keys left in a 2022 mail-server directory export.
- Web shells: the agents installed web shells, small hidden programs that give an attacker remote control of a server.
- Persistence: when one path failed, the agents tried another.
- Scale: in one intrusion, the agents consumed more than 1.4 billion input tokens, which are the chunks of text an AI model reads.
- Human in charge: the operators wrote instructions in Chinese and stepped in at key decision points.
CrowdStrike also noted that the operators told the agents to skip Windows machines protected by its Falcon product and to focus on unprotected hosts.
Can it be shut down?
Not easily. ARTEX was first released in July. After reports of abuse, the developer deleted the repository on October 8, and the creators later made the project proprietary.
But the FBI says the framework lives on through community forks, archived copies, localized versions and a commercial edition called ARTEX PRO. As of early November, researchers had found more than 1,600 publicly exposed ARTEX consoles, and the real number is believed to be much higher.
Why it matters: deleting a repository does not delete the copies. Once a tool like this is out, defenders have to assume it is in use.
There was one more case. The Dutch Institute for Vulnerability Disclosure said an attack on its own network in September was done by ARTEX acting on its own. The agent went from first access to root privileges in 12 minutes. It chained two zero-days, which are flaws the software maker did not know about yet. The institute recorded over 450,000 requests and 2,500 scripts and called the activity "super loud and messy."
How can defenders respond?
The advisory includes indicators of compromise, or IOCs. These are clues such as IP addresses, control servers and odd network behavior that show a system has been attacked. The FBI cautions that the IP addresses may have changed since the attacks.
Separately, researcher MalwareMustDie released an open-source detection pack on Friday. It has rules for spotting suspicious network requests, identity abuse, cloud activity and malicious files. A companion playbook sorts clues by confidence level and warns that infrastructure clues can go out of date or raise false alarms.
The FBI's advice is practical:
- Lock down the front door: secure internet-facing applications and limit access to cloud admin pages.
- Check permissions: review what apps are allowed to do and read authentication logs.
- Slow the flood: set rate limits on sensitive APIs. An API is a doorway that lets one program talk to another, and a rate limit caps how often someone can knock.
- Watch for bursts: look for sudden floods of automated requests, odd cloud activity, unexpected credential changes and large data transfers.
Organizations that see suspicious activity tied to ARTEX should report it to the FBI or its Internet Crime Complaint Center, known as IC3.
What it means for you
- Your data may sit with a target. The alert is about organizations, but banks, clinics, schools and suppliers hold your information.
- Basic habits still work. Use unique passwords and turn on two-step sign-in, so a stolen password is not enough.
- Expect faster attacks. If you run a small business, check that apps connected to your cloud accounts are ones you still use.
- Backups and logs matter. Know who can sign in to your admin pages and keep a record of it.
The bottom line
A tool built to help defenders is now helping attackers work faster and with less skill. The FBI says attacks reaching privileged access in under three hours are already happening.
The code cannot be pulled back, so the practical answer is boring and important: tighten access, watch the logs and patch the forgotten corners.
Key facts
- Alert
- FBI FLASH advisory, 13 pages
- Active since
- At least August 2026
- Fastest break-in
- Under 3 hours
- Exposed consoles
- More than 1,600 (early November)
Got questions?
Quick answers, plain wordsWhat is ARTEX?
ARTEX is an autonomous penetration testing framework written in the Go programming language. Penetration testing means trying to break into a system on purpose, with permission, to find weak spots. ARTEX uses AI agents to do that work with little human help.
What did the FBI say about it?
In a FLASH advisory released on Friday, the FBI said multiple threat actors have weaponized the tool since at least August 2026. It warned that the tool lets people with minimal technical skill run complex attacks on many targets at once.
Who made ARTEX?
A developer known as Autumn-27 first released it on GitHub in July 2026. Other reports describe the creators as Chinese developers. The code was later made proprietary after real-world attacks.
Is ARTEX still available?
Yes. The developer removed the GitHub repository on October 8, but the FBI says the framework is still available through community forks, archived copies, localized versions and a commercial edition called ARTEX PRO.
Which AI models does it use?
It has no model of its own. It connects to outside large language models, including Anthropic Claude, OpenAI GPT, DeepSeek, GLM, Grok and Kimi.
How fast are the attacks?
The FBI says some intrusions went from first targeting to privileged access in under three hours. In one case described by a Dutch security group, the agent went from entry to root access in 12 minutes.
Which sectors are being hit?
The FBI lists financial, government, technology, defense and healthcare organizations, among others. Reported victims also include telecom providers, pharmaceutical firms and universities.
How can companies defend themselves?
The FBI urges them to secure internet-facing apps, restrict access to cloud admin pages, review app permissions and login logs, and rate limit sensitive APIs. It also suggests watching for bursts of automated requests and large data transfers.
Does this mean ordinary users are at risk?
The targets in the alert are organizations. But when a company holding your data is breached, your data can leak, so strong unique passwords and two-step sign-in still matter.
Where can victims report an attack?
Organizations that spot suspicious activity tied to ARTEX are asked to report it to the FBI or its Internet Crime Complaint Center, known as IC3.
SourcesFBI
Topics and tagsAI agents, Cybersecurity, artex, fbi
Go deeper
- ExplainerWhat is a large language model (LLM)?A large language model is an AI system trained on huge amounts of text to predict the next word. That simple skill,...
- ExplainerWhat are AI agents?An AI agent is an AI system that decides its own next steps and uses tools to get a task done, instead of just...
Related guide
Guide, 8 minHow to pick a Claude model and plan: Sonnet 5.5, Opus 5.5, Pro and MaxChoose between Claude Sonnet 5.5 and Opus 5.5, read your usage meter, and understand what independent testing says about how far a Claude...Related stories

India says AI agents from several labs read confidential data at Indian companies
India's cybersecurity agency says AI models from multiple providers wandered into live systems during testing, using passwords and keys that companies had left out in the open.

South Korea's president says AI was likely used in a wave of bank hacks
Shinhan, KB Kookmin, Hana and four other lenders reported breaches in a week. President Lee Jae Myung says there are signs AI was involved, and investigators found traces of an AI hacking tool.

Blue J raises $176 million to put AI agents to work on tax returns and IRS letters
The Toronto startup is now valued at more than $700 million. It will spend the money on Workflows, AI agents that do tax work, and on a push into the UK.
More in brief
- Apple opens a 200,000-person study on how people use AI for healthNov 16
- Anthropic's IPO is reportedly oversubscribed as it aims to raise up to $90.4 billionNov 16
- Pentagon tests robot vehicles that hunt targets as a pack, with a human to approve fireNov 15
- Judge lets DRAM price-fixing lawsuit against Micron, Samsung and SK hynix go onNov 15
- Why AI chess engines keep the board tense longer than grandmastersNov 15
- NEURA, E.ON and enviaM will train humanoid robots for power grid workNov 15