All articles
ArticlesNovember 16, 20267 min read

FBI ARTEX alert: how to defend against AI-driven attacks

The FBI ARTEX alert warns AI agents now run intrusions in under three hours. What ARTEX does, known campaigns, and the defenses to put in place first.

fbi artex alertartexai agentsagentic attacksthreat intelligencecloud securityapplication securityfbi
FBI ARTEX alert: how to defend against AI-driven attacks

The FBI published a FLASH alert on November 13, 2026 about ARTEX, an open-source AI agent framework that attackers now use to run intrusions almost end to end. The FBI ARTEX alert (FLASH-20261113-001) says some intrusions went from first targeting to privileged access in under three hours, and that people with little technical skill are running multi-target campaigns with it.

I work on security at erxes, and most of my week goes to the boring parts of an attack surface: partner portals, old APIs, admin panels somebody exposed for a demo. That is exactly where the FBI says ARTEX does its best work. So what worries me here is less the tool itself and more how fast the usual mistakes now get found and chained.

What the FBI ARTEX alert says

According to the FBI FLASH on ic3.gov, ARTEX turns goals written in plain language into commands that specialized agents execute, with very little human input after the objective is set. It ships no model of its own. It calls out to whatever LLM provider the operator configures.

A few details from the alert stood out to me:

  • In attacker-controlled instances, Chinese models were the most common, and DeepSeek Flash family models dominated. The FBI points to their low cost and the lack of observable guardrails when a request is framed as penetration testing.
  • A developer using the handle "Autumn-27" released ARTEX on GitHub in July 2026. It picked up an estimated 1,500 to 2,000 stars within 10 weeks.
  • After reports of abuse, the developer made the repository unavailable on October 8, 2026. The FBI says the framework keeps spreading through community forks, localized versions, archived copies and a commercial edition called ARTEX PRO.
  • Researchers counted more than 1,600 publicly exposed ARTEX consoles as of early November, and the FBI expects the real number of deployments to be much higher.

Targets so far span financial services, government, technology, defense, semiconductors, energy, healthcare, manufacturing, telecom and legal, going back to at least August 2026.

Known campaigns using ARTEX

The alert links ARTEX to a few campaigns that were already public. BleepingComputer's write-up covers them in more detail:

  • ConnectWise Cloud Backup: Microsoft attributed this supply chain attack to a China-aligned actor it tracks as Storm-0940. After compromising the application identities used by the backup service, the attackers used ARTEX to try authenticating against more than 2,250 downstream Microsoft 365 tenants.
  • South Korean banks: A Chinese-speaking actor paired ARTEX with Claude Code agents against several banks, and the FBI says personal data of roughly 67,000 people was exfiltrated. CrowdStrike found the attacker's exposed infrastructure, including ARTEX config files and Claude Code session histories.
  • Vapor Panda: CrowdStrike reported this espionage group used ARTEX against more than 200 organizations in 14 countries and compromised at least nine.

How ARTEX gets in

The TTP table in the alert reads like a list of web app findings I've seen in real reviews, now run by agents that don't get bored. Initial access included:

  • unsigned SAML assertions and forged SAML tokens
  • XXE, SSRF, SQL injection, command injection and SSTI leading to RCE
  • IDOR and web APIs with no authentication at all
  • hardcoded admin passwords sitting in single-page app JavaScript bundles
  • weak admin credentials, stolen VPN configs and leaked credentials
  • abuse of cross-account cloud trust and compromised vendor credentials over B2B federation

On the cloud side, agents mapped the effective permissions of stolen Azure credentials (SAS tokens, service principals, Kudu credentials, storage keys) with sub-second capability checks, test blob writes and ARM deployment dry runs. For persistence they dropped JSP and PHP web shells, abused Kudu admin APIs, added credentials to existing OAuth apps, and installed open-source RATs like WISP, vshell and AdaptixC2.

None of these techniques is new. The change is that an agent tries them across hundreds of peripheral systems at once and keeps going when the first path fails.

How to defend against ARTEX attacks

The FBI's recommendations are practical. Here is how I'd turn them into a short checklist for a small team or a SaaS product like ours.

Shrink the attack surface you forgot about

Make a real inventory of internet-facing apps, including partner and supplier portals, mobile backends, employee tools, sales-support systems, legacy services and integrations other teams built. Put cloud admin consoles, Kudu/SCM endpoints, backup platforms and hypervisors behind trusted networks or private access. The FBI also suggests requiring an architecture review before any workload or management interface gets a public IP.

Treat app permissions as credentials

The ConnectWise case worked because a third-party app already had broad consent in customer tenants. Apply least privilege to multi-tenant app permissions, watch for new credentials or privilege grants on high-privilege apps, and review service principal sign-ins (successful and failed separately) against your vendor's expected infrastructure, not just against published IOCs.

Put limits on business APIs

This is the one I'd prioritize for anyone running a CRM or ERP. The FBI recommends rate limits, recipient limits and transaction thresholds on high-impact APIs, alerts on sequential lookups or unusually many records per session on portals, server-side authorization on every customer record, and responses trimmed to the fields an operation actually needs. Agents enumerate. Make enumeration loud and slow.

Stop leaking the map

Don't return stack traces, SQL errors, internal paths or framework details to unauthenticated users. And scan configs, environment variables, infrastructure-as-code, CI/CD pipelines, container images and repos for secrets continuously, then rotate whatever turns up. I wrote about how one leaked vendor key spiraled in the Cursor security incident, and ARTEX is the kind of tool that finds such keys for you.

Watch for agent-shaped traffic

The alert lists behavioral signs: bursts of high-frequency requests with sequential enumeration against login and search endpoints, many failed attempts with small incremental changes, duplicate requests from sub-agents stepping on each other, and unexpected outbound calls to LLM APIs like api.anthropic.com, api.deepseek.com, open.bigmodel.cn or api.x.ai from servers that have no reason to make them. Self-hosted LLM gateways or reseller software in your egress logs are worth a look too.

Why this matters for teams building agents

It's uncomfortable, but ARTEX is a well-built agent system. It plans, splits work across sub-agents, retries and adapts, the same patterns we use for coding and support agents. The difference is the goal. If you run your own agents, the controls I covered in securing AI agents with Uber's framework apply both ways: scoped credentials, logged tool calls and hard limits on what an agent can touch.

For defenders, the attack window shrinks to hours. Quarterly reviews of exposed assets and app consents won't keep up with that. Detection and rate limiting on boring endpoints will matter more than any single IOC list, especially since the FBI itself warns that the listed IPs may have been reassigned.

FAQ

What is ARTEX?

ARTEX is an open-source, autonomous multi-agent penetration testing framework written in Go. It connects to external LLMs and uses agents to run reconnaissance, exploitation, privilege escalation and data exfiltration based on goals written in natural language.

Is ARTEX still available after the GitHub repo was removed?

Yes. The FBI says the original repository went offline on October 8, 2026, but forks, localized variants, archived copies and a commercial ARTEX PRO version keep it available.

What should I do first after reading the FBI ARTEX alert?

Inventory your internet-facing apps, including peripheral ones, restrict admin and deployment interfaces to trusted networks, review third-party app permissions and service principal sign-ins, and add rate limits to sensitive APIs. Check the alert's IOCs against your logs, but treat IP indicators as historical.

Are there detection rules for ARTEX?

BleepingComputer reports that researcher MalwareMustDie released an open-source ARTEX detection signature pack and playbook on November 13, 2026, which sorts indicators by confidence level.


Sources

Cover image: Seal of the Federal Bureau of Investigation, Wikimedia Commons (public domain)

Related articles