ARTEX AI hacking tool: FBI says attacks take under 3 hours
On 13 November 2026, the FBI warned that ARTEX, an AI attack tool, took some intrusions to admin access in under three hours. What it targets and what to check.
On 13 November 2026, the FBI published a FLASH alert about ARTEX, an AI tool that runs most steps of a cyberattack by itself. In some intrusions, ARTEX went from the first scan to privileged access in under three hours. On the same day, Amazon published its own report on attackers who used ARTEX.
The FBI says that ARTEX lets attackers "with minimal technical expertise" run complex attacks against many targets at the same time. The entry points in the alert are mostly basic ones: weak passwords, keys left in code and pages without a login.
What ARTEX is and who uses it
ARTEX is an open-source penetration testing framework. Security teams use this type of tool to test their own networks. The user writes a goal in plain language, and AI agents turn it into commands. ARTEX has no AI model of its own. It connects to models from Anthropic, OpenAI and Chinese providers such as DeepSeek.
According to the FBI, a developer called "Autumn-27" released ARTEX on GitHub in July 2026. Since at least August 2026, attackers have used it against financial services, government, technology, healthcare, manufacturing and legal organizations. The developer removed the repository on 8 October 2026. Copies, forks and a paid version, ARTEX PRO, are still available. The FBI says that in early November researchers found more than 1,600 ARTEX consoles open on the internet.
The FBI names two campaigns:
- Attacks on South Korean financial institutions took personal data of about 67,000 people.
- An attack on ConnectWise Cloud Backup customers gave access to more than 2,250 downstream Microsoft 365 tenants.
ARTEX attacks the systems that get less attention
The FBI says that ARTEX "excels" at peripheral systems: partner and supplier portals, employee apps, legacy services and third-party integrations. The ways in that the alert lists include:
- administrator passwords written into the JavaScript code of a web page
- web APIs with no login
- weak administrator passwords
- stolen VPN settings and leaked passwords
- passwords of a third-party vendor
TechCrunch reports that the South Korean attacks worked mostly because of weak defenses and no multi-factor authentication. According to the FBI, ARTEX makes a lot of network traffic that can set off alerts, but the attacks are so fast that defenders have little time to react.
A vendor's old key opened Barracuda's cloud
The security company Barracuda published its own incident report on 11 November. On 1 November 2026, an attacker logged in with a valid cloud access key. Barracuda had given that key to a third-party AI vendor for an earlier project. Nobody removed the key when the project ended. The vendor's development systems were hacked in October, and the attacker probably found the key there.
The attacker used each key that it found to open the next system. It reached about 10 cloud accounts, source code access and Kubernetes clusters. The activity stopped in under two days, before an engineer reported it on 4 November. Barracuda links the method to ARTEX but does not say who the attacker was.
Amazon saw AI agents steal keys for AI tools
Amazon's report, by Amazon's chief information security officer CJ Moses, covers 3 October to 5 November 2026. Amazon assesses with moderate confidence that the attackers used ARTEX. The agents stole API keys for AI models and cloud passwords. They looked for self-hosted AI tools such as Dify, Langflow, Open WebUI, Ollama and LiteLLM. They used default administrator passwords and secrets left in web page code.
Amazon could not confirm the results that the agents reported. Its advice is to fix default passwords, keys in code, interfaces without a login and missing updates first: "Getting those fundamentals right remains the most effective defense." Stolen AI keys can also cost money directly, as our article on LLMjacking shows.
What to do this week
- 01List every system that your business has on the internet.Include supplier portals, booking and staff apps, old websites and AI tools that someone set up for a test.
- 02Turn on multi-factor authentication for email, cloud, VPN and every admin login.TechCrunch names missing multi-factor authentication as a cause of the South Korean attacks.
- 03Remove the keys and passwords of every past vendor and finished project.Barracuda's attacker used a key that a vendor kept after the work ended.
- 04Change every default administrator password, also on self-hosted AI tools.Ask your developer to search your code and website files for keys and passwords, and to replace any that they find.
- 05Put admin pages behind a VPN or a trusted network, so they do not open from the public internet.This is the first group of recommendations in the FBI alert.