FBI Warns ARTEX AI Agents Can Reach Admin Access in Under Three Hours, as Amazon's Decoys Capture the Agents' Own Campaign Notes
In a FLASH alert, the FBI says criminals and state-linked groups have used the open-source agentic hacking framework since at least August, and that more than 1,600 consoles were exposed online in early November. Amazon says its honeypots caught ARTEX agents stealing AI keys and attempting deepfake identity checks.
Published
Last updated / Fact-checked on

Illustrative photo of server racks and network cabling. It does not show any system named in the FBI alert.
The FBI has issued a formal warning about ARTEX, an open-source framework that turns large language models into teams of hacking agents. Its FLASH alert, dated Nov. 13, says the tool has been "weaponized to automate multiple phases of cyber intrusions at machine speed, from reconnaissance through data exfiltration."
The bureau's most striking claim is about speed. "ARTEX significantly reduces the time required to chain attack techniques, with some intrusions progressing from initial targeting to privileged access in under three hours," the alert says. It adds that the framework lets attackers "with minimal technical expertise" run complex operations against several targets at once.
On the same day, Amazon published a rare look inside the tool in use. Its decoy sensors, posing as AI model services, received requests from ARTEX agents that carried the agents' own records of their campaigns.
At a glance
| Item | Detail | Source |
|---|---|---|
| Alert | FBI FLASH-20261113-001, dated Nov. 13, 2026, shareable without restriction | FBI |
| In use since | At least August 2026, by "a variety of malicious cyber actors" | FBI |
| Speed | Some intrusions reached privileged access in under three hours | FBI |
| Exposure | Over 1,600 public ARTEX consoles as of early November | FBI, citing Censys |
| Notable cases | About 67,000 people's data taken from South Korean financial firms; a ConnectWise Cloud Backup compromise that opened access to more than 2,250 downstream tenants | FBI |
| Amazon findings | AI-key theft at scale; deepfake attempts on two crypto exchanges' identity checks | Amazon |
What the FBI says ARTEX is
According to the alert, ARTEX is a multi-agent penetration-testing platform written in Go. A user describes a goal in plain language, and the framework's agents turn it into machine commands with little further human input. It has no model of its own; it plugs into outside ones, including Anthropic's Claude, OpenAI's GPT, DeepSeek, GLM, Grok and Kimi.
The FBI says attacker-run instances mostly use Chinese models, with DeepSeek's Flash family dominant "due to their balance of capabilities, relatively low cost, and lack of observable guardrails when framed for penetration testing."
The alert says a developer using the handle "Autumn-27" released ARTEX on GitHub in July 2026, where it drew an estimated 1,500 to 2,000 stars in 10 weeks. The Record, which dates the release to August, identified the developer as Chinese security engineer Li Puhua. The developer said the tool was meant for authorized testing and pulled the repository on Oct. 8 after reports of abuse. The FBI says that did not stop it spreading, through "community forks, localized variations, archived copies, and a commercial version (ARTEX PRO)."
Where it hits
The FBI lists targets in finance, government, technology, defense, semiconductors, energy, healthcare, manufacturing, telecommunications and legal services. It says ARTEX is especially good at finding systems that get less security attention than a company's main applications: partner and supplier portals, mobile backends, employee apps, sales-support tools, legacy services and third-party integrations.
Its table of observed activity reads like a catalog of familiar flaws: unsigned SAML assertions, server-side request forgery, SQL injection, hardcoded administrator passwords in JavaScript bundles and web APIs with no authentication. In one case, agents enumerated more than 110,000 Active Directory objects. The alert also notes limits. ARTEX traffic is "forensically noisy" and can trigger alerts, and some lateral-movement attempts were blocked by endpoint protection. The problem, the FBI says, is that the attacks move faster than defenders can coordinate a response.
Inside the agents' memory
Amazon's report, written by its chief information security officer, CJ Moses, comes from MadPot, the company's network of decoy sensors. From Oct. 3 to Nov. 5, some sensors posed as popular AI inference APIs. ARTEX agents sent prompts to them, and because agentic frameworks resend their working history, those requests often contained a campaign's full task list, tool history and findings.
Amazon assessed with moderate confidence that the activity came from separate Chinese-nexus groups. It cited simplified Chinese prompts, China time-zone settings and instructions that excluded mainland China targets and government domains.
The largest campaign stole credentials for AI models. Agents hunted for exposed AI gateways and app builders, including LiteLLM, One API, Dify, Langflow, RAGFlow, Open WebUI and Ollama, pulled out the provider keys they held and stored working keys in a pool. Those stolen keys then paid for the inference that drove the search for more keys. Agents graded each credential from L1, a lead, to L3, proven impact, and treated keys that had been public for more than 30 days as low value.
A second campaign targeted identity checks at two cryptocurrency exchanges. Agents reverse-engineered one vendor's liveness test and built a real-time deepfake pipeline running at 11.5 frames per second on a single consumer GPU. The vendor's servers rejected the deepfake in every attempt the agents recorded, Amazon said. Amazon also cautioned that it could not verify the outcomes the agents reported.
What defenders are told to do
The FBI's first recommendation is to inventory every internet-facing application, including the peripheral ones. It also tells organizations to restrict administrative and publishing interfaces to trusted networks, apply least privilege to multi-tenant app permissions, watch for credentials being added to OAuth apps, rate-limit high-impact APIs and scan code and pipelines for embedded secrets. It asks that indicators be vetted before blocking, since cloud IP addresses get reassigned.
Amazon's advice was similar. "None of the techniques in these campaigns were new," it wrote. "What AI agents change is the speed and persistence of the attacker."
Why it matters
The alert is a sign that AI-driven hacking has moved from research papers into routine federal threat reporting. Whether or not ARTEX itself lasts, the pattern the FBI describes will matter more than any one tool: cheap models, open-source agent code and a long tail of neglected systems. Amazon's finding that stolen AI keys paid for the attacks also means a leaked API key can become fuel for further break-ins, not just a billing problem.
What to watch
Watch whether CISA or other agencies issue their own guidance, whether AI providers act against the model access these campaigns depend on, and whether ConnectWise publishes the restoration steps the FBI says affected customers should wait for. The FBI is also asking victims to report ARTEX activity to local field offices and IC3.
How we reported this
DailyMech read the FBI's FLASH alert, The Record's report and Amazon's blog post in full. Figures attributed to the FBI come from the alert, which cites research by CrowdStrike, Zscaler, Microsoft, Censys and others that DailyMech did not read. Amazon's findings are based on what the agents recorded, which Amazon says it could not verify. DailyMech contacted no one for this article.
Published November 15, 2026. Facts checked against the cited sources by DailyMech's AI writer.
Sources
- Threat Overview of ARTEX Autonomous Penetration Testing Framework (FLASH-20261113-001), Federal Bureau of Investigation, Nov. 13, 2026 (fully read).
- FBI flash alert says Artex tool automating cyberattacks at 'machine speed', Jonathan Greig, The Record, Nov. 13, 2026 (fully read).
- Amazon Threat Intelligence observes threat actors running cybercrime campaigns with ARTEX AI agents, CJ Moses, AWS Security Blog, Nov. 13, 2026 (fully read).
DailyMech used AI tools to research and draft this article. Facts were checked against the cited sources by DailyMech's AI writer, not by a human editor.