FBI says AI tool ARTEX sped some intrusions to admin access in hours
An FBI alert says the open-source framework has been used against at least ten sectors since August. Barracuda and ConnectWise are among the victims named in reports.
Key takeaways
- An FBI FLASH alert says several groups have used the open-source AI framework ARTEX in attacks since at least August 2026.
- The FBI says some intrusions reached privileged access less than three hours after the first targeting.
- ARTEX plugs into outside AI models; the FBI says Chinese models, led by DeepSeek's Flash models, are most common on attackers' servers.
- The alert links the tool to the ConnectWise supply-chain attack and to breaches at South Korean banks; Barracuda has disclosed an ARTEX-linked breach.
- The original GitHub repository came down on Oct. 8, but forks, copies and a paid edition keep the tool available, the FBI says.
The FBI has warned that several groups of hackers are using ARTEX, an open-source framework that lets AI agents carry out most stages of a cyberattack, against organisations around the world, according to BleepingComputer, The Record, TechCrunch and The Next Web. In a FLASH alert published on Friday, Nov. 13, the bureau said the tool has been used since at least August 2026, BleepingComputer reports.
The FBI wrote that ARTEX "significantly reduces the time required to chain attack techniques," and said some intrusions moved from the first targeting of a victim to privileged access in less than three hours, according to BleepingComputer and The Record. The bureau also said the framework lets attackers with little technical skill run complex operations against many targets at once.
What is ARTEX?
ARTEX was built as a penetration testing tool, the kind of software security teams use to probe their own defences. BleepingComputer describes it as written in Go, with AI agents handling reconnaissance, finding and exploiting flaws, raising privileges and taking data while people do little. Rather than shipping a model, it connects to outside large language models, including Anthropic's Claude, OpenAI's GPT and Chinese-origin models such as DeepSeek, GLM and Kimi, the outlets report.
According to The Record, the FBI said Chinese models are the most common on attacker-controlled ARTEX servers, with DeepSeek's Flash models dominating because they are relatively cheap and show no observable guardrails when a request is framed as penetration testing.
A developer known as Autumn-27 published the tool on GitHub, in July according to BleepingComputer, TechCrunch and The Next Web, or in August according to The Record. The Record identifies the developer as Chinese cybersecurity engineer Li Puhua and says the developer denied responsibility for crimes committed with the tool. The developer had said the tool was meant only for authorised testing, The Next Web reports. The repository was taken down on Oct. 8 after reports of abuse, BleepingComputer and The Next Web report; TechCrunch says the developer made the software closed-source. The FBI says forks, archived copies, localised versions and a paid edition called ARTEX PRO keep it in circulation, and by early November more than 1,600 ARTEX consoles were found publicly exposed, a count The Record attributes to the security firm Censys.
Which attacks has it been linked to?
The FBI's alert ties ARTEX to the October supply-chain attack on ConnectWise's cloud backup service, which Microsoft attributed to a China-state-aligned group it tracks as Storm-0940, BleepingComputer reports. The attackers used the tool to try to log in to more than 2,250 downstream Microsoft 365 tenants, according to BleepingComputer, which also cites an earlier Huntress count of at least 162 compromised tenants. TechCrunch instead describes the 2,250 as customers whose Microsoft 365 environments the hackers accessed. ConnectWise did not respond to TechCrunch's request for comment on the FBI's claim.
The alert also links the tool to attacks on South Korean banks that exposed personal data on about 67,000 people, according to BleepingComputer and The Next Web; TechCrunch puts the figure at least 68,000. In that campaign, a Chinese-speaking attacker combined ARTEX with Anthropic's Claude Code agents, BleepingComputer reports, citing CrowdStrike research. TechCrunch reports that those attacks were blamed largely on weak defences, such as missing multi-factor authentication. The Record says the tool was also tied to attacks on Japanese financial institutions.
Barracuda, a security company, said attackers used ARTEX and a cloud access key it had given a third-party AI vendor for an earlier project, and had not retired once that work ended, BleepingComputer reports. The attacker reached about 10 of its cloud accounts over 44 hours on Nov. 1 and 2, the company said. Accounts of its response differ. BleepingComputer reports Barracuda found no evidence that stored customer integration credentials were taken, but asked potentially affected customers of two of its services to revoke and recreate them. TechCrunch reports that Barracuda said its logs were not detailed enough to rule out access and that it changed all of its customer credentials; a spokesperson later told TechCrunch it had directly notified a minority of its XDR customers.
What else did researchers find?
Amazon released a separate study the same day the alert came out, The Record and The Next Web report. Its decoy sensors caught ARTEX agents in a campaign that stole access keys for AI models and used them to pay for the AI that searched for more keys, The Next Web reports. Amazon's chief information security officer, CJ Moses, wrote that the agents got in where basic security practices were lacking, such as default credentials, keys left in code and unpatched software, according to The Next Web.
What does the FBI advise?
The FBI urges organisations to lock down their public-facing apps, limit who can reach cloud admin consoles, review permissions and login records, and put rate limits on sensitive APIs, BleepingComputer reports. It also asks organisations to keep admin interfaces off the open internet and to list every system they expose online, including partner portals and employee apps, The Next Web reports. The bureau notes that IP addresses in its list of indicators may have changed since the attacks.
Disclosure: NextFish's AI drafting desk uses Claude, a model made by Anthropic.
Questions readers are asking
What is ARTEX?
An open-source penetration testing framework in which AI agents carry out reconnaissance, exploitation and data theft with little human input. It connects to outside AI models rather than using its own.
How fast are ARTEX attacks?
The FBI says some intrusions reached privileged access less than three hours after a victim was first targeted.
What should organisations do?
The FBI advises securing internet-facing apps, keeping cloud admin interfaces off the open internet, reviewing permissions and logs, rate-limiting sensitive APIs and listing every system exposed online.
Sources
- BleepingComputer: FBI warns ARTEX AI hacking tool is powering global cyberattacks
- The Record: FBI flash alert says Artex tool automating cyberattacks at 'machine speed'
- TechCrunch: FBI says Artex AI open source hacking tool used to breach networks in 'under three hours'
- The Next Web: FBI says AI hacking tool ARTEX can break into networks in under three hours
- BleepingComputer: Barracuda links cloud breach to ARTEX AI-powered cyberattack
- #artex
- #fbi
- #ai-agents
- #cyberattacks
- #penetration-testing
Get the day's top AI & tech stories on WhatsApp: Follow the NextFish channel
Get stories like this every morning.
Related
Cybersecurity3 min
Optimizely flaw let attackers plant malware on BBC, CNN and Zoom sites
A researcher found malicious code in scripts used by 178 organisations. Optimizely says it fixed the flaw, removed the code and notified affected customers.
Cybersecurity2 min
Google Project Zero will cut its bug disclosure deadline to 30 days
The team says AI has sped up how fast attackers find and exploit flaws. It drops its extra 30-day patch 'soak time' now and moves to 60 days in 2027 and 30 days in 2028.
Cybersecurity4 min
ConnectWise urges ScreenConnect client updates as SCKit targets technicians
Microsoft, Huntress and Blackpoint have analysed the toolkit used against MSPs. It is built to jump to technicians, but researchers say they have not seen that work.
Cybersecurity3 min
Landmark Group says Shukran loyalty members' personal data was accessed
Names, contact details and dates of birth were involved, the Dubai retailer says, but not card details, passwords or points. It has not said how many members were affected.