ARTEX: The FBI Warns Agentic AI Has Turned Pentesting Into Machine-Speed Intrusion
Bit4Bit DFIR, practical notes for people who actually run the systems.
This one is not a CVE, and there is no patch to install. On November 13, 2026, the FBI published FLASH-20261113-001, "Threat Overview of ARTEX Autonomous Penetration Testing Framework," warning that threat actors have weaponized an open-source, LLM-driven agentic platform to automate intrusions from reconnaissance through data exfiltration, at machine speed.
The headline number to carry into your next hunt: some ARTEX-driven intrusions went from initial targeting to privileged access in under three hours. That is not a faster scanner. That is a compressed detection window, and it changes what your logs need to show you.
What ARTEX is
ARTEX is an autonomous, multi-agent penetration testing platform written in Go. An operator states a goal in natural language, and the framework translates it into commands carried out by specialized agents, with minimal human interaction after the objective is set. It ships with a web console, and it has no model of its own. It connects to external LLM providers, including Anthropic Claude, OpenAI GPT, DeepSeek, GLM, Grok, and Kimi.
Analysis of attacker-controlled instances found Chinese models most prevalent, with DeepSeek Flash family models dominating, which the FBI attributes to their balance of capability, relatively low cost, and lack of observable guardrails when framed for penetration testing.
The developer, "Autumn-27," released ARTEX on GitHub in July 2026. It gained 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. That did not kill it. The FBI says ARTEX remains available through community forks, localized variations, archived copies, and a commercial version, ARTEX PRO. As of early November 2026, researchers identified over 1,600 publicly exposed ARTEX consoles, with total deployments likely significantly higher once systems behind firewalls or in private networks are counted.
Who has been hit, and how
The FBI says a variety of malicious actors have used ARTEX since at least August 2026, targeting organizations globally across financial services, government, technology, defense, semiconductor, energy, healthcare, manufacturing, telecommunications, and legal industries.
Two campaigns in the FLASH are worth knowing by name:
- South Korean financial institutions. Financially motivated attacks exfiltrated personal information belonging to approximately 67,000 individuals.
- ConnectWise Cloud Backup supply chain. A compromise affecting ConnectWise Cloud Backup customers enabled access to more than 2,250 downstream tenants. Microsoft attributed related ARTEX activity to Storm-0940, described as a China-state-aligned actor.
ARTEX excels at the systems most teams watch least. The FBI calls out peripheral systems receiving less security attention than flagship applications: partner and supplier portals, mobile backends, employee applications, sales-support systems, legacy services, third-party integrations, and applications created across different business units. Agents systematically map trust relationships, then pivot from a weaker third-party vendor, CI/CD pipeline, or integrated provider into a higher-value upstream network.
The observed tactics are familiar techniques chained at unfamiliar speed. Entry included unsigned SAML assertions, XXE, SSRF, anonymous file upload, missing authentication on web APIs, hardcoded administrator passwords in JavaScript bundles, and remote code execution through command injection, SSTI, and SQL injection. Once in, agents deployed JSP and PHP web shells, abused Kudu administrative APIs, created SSH key pairs, added credentials to existing OAuth applications, ran Kerberoasting and DCSync, and enumerated over 110,000 Active Directory objects in one observed case, including reading LAPS passwords and inventorying endpoint sensor coverage to find machines without CrowdStrike Falcon.
In cloud environments, the tell is sub-second: ARTEX mapped effective permissions of Azure credentials using SAS tokens, service principals, and storage account keys, with capability checks, test blob writes and deletes, and ARM deployment validation dry runs all happening in under a second per check.
What to do
There is no version to upgrade to. The mitigation is exposure and identity hygiene, and the FBI's recommendations start with the inventory most teams have never finished.
Reduce the external attack surface:
- Inventory all internet-facing applications, including the peripheral ones: partner and supplier portals, mobile backends, employee applications, sales-support systems, legacy services, and third-party integrations.
- Restrict cloud administration, network device management, application publishing interfaces such as Kudu/SCM endpoints, hypervisors, and backup platforms to trusted networks, private connectivity, or dedicated administrative paths. Disable basic publishing authentication where it is not required.
- Move enterprise applications behind Zero Trust access solutions to eliminate external exposure where feasible.
Harden identity and access:
- Apply least privilege to multi-tenant application permissions and consent grants, and monitor credential additions, privilege assignments, and anomalous app-only activity.
- Review successful and failed service-principal sign-ins separately. Investigate unfamiliar sources and activity outside expected vendor infrastructure, and do not limit the review to published IP addresses.
- Review Microsoft Entra audit records for application and service principal credential additions, new applications and owners, app-role assignments, consent grants, and privileged role assignments. If suspicious activity is confirmed, revoke unnecessary permissions and rotate credentials.
- If your organization used ConnectWise applications, review every Microsoft 365 tenant where they are or were installed or authorized, even if the integration is no longer in use, and search by application ID rather than display name.
Strengthen application protections:
- Put guardrails around high-impact business APIs: rate limits, recipient limits, transaction thresholds, and abuse controls appropriate to the function. Alert when an application operates materially outside normal volume.
- On partner and employee portals, alert on sequential lookups and unusually high numbers of records accessed per session, apply rate limits, and enforce server-side authorization for access to customer records.
- Scan repositories, container images, build artifacts, and CI/CD pipelines for embedded secrets, and rotate exposed credentials promptly.
The DFIR angle
ARTEX generates high volumes of forensically noisy traffic that may trigger alerts, but the FBI is explicit that attack speed leaves limited time for coordinated defender response before objective completion. Your hunt should look for speed and iteration, not stealth.
Start with these patterns from the FLASH:
- Compressed bursts: high volumes of failed attempts with iterative, incremental changes, and multiple sub-agents overwriting each other or generating duplicate requests.
- Cloud capability probing: Azure ARM deployment validation dry runs across multiple resource types, test blob write and delete operations with sub-second timing, cloud metadata service queries, and unexpected IAM role assumptions or PassRole operations.
- LLM egress: outbound connections to LLM API endpoints, unfamiliar LLM proxies or resellers in egress logs, and token consumption or API billing anomalies. The FLASH names api.anthropic.com, api.deepseek.com, open.bigmodel.cn, and api.x.ai as endpoints to investigate alongside other suspicious activity, and xcai.pro as an LLM API reseller seen in the South Korean campaign.
- Persistence and staging: JSP or PHP web shells in unexpected paths, newly created accounts, SSH key additions, Elastic IP allocations, unexpected EC2 or Azure VM launches, new 100 GB volumes created for exfiltration preparation, and command output sent to rotating webhook.site endpoints.
- Identity artifacts: Kudu/SCM credential abuse, OAuth application credential modifications, and service-principal sign-ins from previously unseen infrastructure, ASNs, hosting providers, or geographies.
One indicator to treat with care: 38.244.50.120, identified by CrowdStrike and Censys as an ARTEX host and open directory in the South Korean financial campaign. The FBI recommends vetting IOCs before blocking on them, because infrastructure used for malicious activity may be hosted on cloud or virtualized platforms where IP addresses can be dynamically or temporarily assigned, and an address associated with malicious activity at one point may later be reassigned to benign services. Corroborate with your own telemetry before you block.
For scope, the FBI recommends reviewing activity starting from at least October 1, 2026, for ConnectWise-related activity, and extending service-principal sign-in review to include failed authentications and volume anomalies. Preserve Entra sign-in and audit records, Microsoft Purview Audit records, SIEM data, and previously configured Microsoft Graph telemetry, and do not delay urgent containment while waiting for exports.
The real lesson of ARTEX is not that AI can hack. It is that the techniques in this FLASH are all old: SAML forgery, SSRF, XXE, hardcoded credentials, missing authentication on an API, a web shell, DCSync. What is new is an agent chaining them in under three hours, against the portal nobody owned, while your team was still triaging the first alert. If your inventory does not include that portal, your detection window is already gone.
Sources: FBI FLASH-20261113-001, "Threat Overview of ARTEX Autonomous Penetration Testing Framework," November 13, 2026, TLP:CLEAR, https://www.ic3.gov/CSA/2026/261113.pdf. Campaign and sector details, observed tactics, indicators, and recommendations are from the FLASH itself, which draws on reports from Microsoft, Google Threat Intelligence Group and Mandiant, CrowdStrike, Zscaler ThreatLabz, Censys, Akamai, Check Point Research, and CERT-EU. The South Korean campaign figure of approximately 67,000 individuals and the ConnectWise figure of more than 2,250 downstream tenants are from the FLASH summary. ARTEX is a framework, not a vulnerability, so there is no CVE number, CVSS score, or patched version to cite.
Top comments (0)