{"slug": "censys-1465-internet-exposed-artex-consoles-many-with-open-directories-and-dual", "title": "Censys: 1,465 Internet-Exposed ARTEX Consoles — Many With Open Directories and Dual-Use Tools", "summary": "Censys ARC principal researcher Silas Cutler published findings on November 5, 2026 identifying 1,465 internet-exposed ARTEX consoles, roughly 815 of which look distinct after removing two cloned hosting clusters, with 155 hosts (10.6%) carrying another Censys threat label and 32 showing public file listings at the time of writing. Censys matched hosts to the ARTEX operator UI using page titles plus Next.js static-asset fingerprints, finding concentration in the United States, Hong Kong, mainland China, Japan, Singapore and South Korea, with PEG Tech (AS54600) accounting for 493 hosts and Yancy Limited (AS138415) for 157. Microsoft's Storm-0940 blog ties ARTEX, an open-source Go backend with a Next.js console that chains LLM planner and worker agents through autonomous penetration testing, to OAuth application abuse following the ConnectWise Cloud Backup incident, meaning tenants may need a service-principal review even without ARTEX installed.", "body_md": "Internet-facing ARTEX consoles are not a hypothesis anymore. On November 5, 2026, Censys ARC principal researcher Silas Cutler published [ARTEX: Agentic Asbestos](https://censys.com/blog/artex-agentic-asbestos/?ref=explainx), counting 1,465 hosts whose HTTP surface matches an ARTEX operator UI fingerprint, mapping where they sit, what shares their IP, and how often attackers leave file listings wide open. TechCrunch [summarized](https://techcrunch.com/2026/11/06/hundreds-of-artex-ai-hacking-tools-found-publicly-exposed-to-the-web/?ref=explainx) the work for a general audience the next day.\n\nThis article is the practitioner reading: numbers, co-location patterns, hunt queries, and identity lessons from Microsoft’s [Storm-0940 blog](https://www.microsoft.com/en-us/security/blog/2026/10/21/storm-0940-uses-artex-agentic-ai-connectwise-supply-chain-compromise/?ref=explainx) — not panic about AI models, but honest risk from misconfigured offensive automation and stolen OAuth trust.\n\n## TL;DR\n\n| Topic | Censys finding | \n|---|---|\n| Hosts | 1,465 ARTEX consoles on the public internet | \n| Distinct-looking | ~815 after removing two cloned hosting clusters | \n| Co-threats | 155 hosts (10.6%) with another Censys threat label | \n| Open dirs | 32 at time of writing with public file listings | \n| Top ports | 8787 (default), 8790, 443, 80 | \n| LLM proxies | 43 hosts with CLIProxyAPI / New API style gateways | \n\nAgents that act on your behalf need rules. [See how explainx helps secure your agents →](https://explainx.ai/agentbeam?src=blog_note)\n\n## What ARTEX is (in one paragraph)\n\nARTEX is an open-source Go backend with a Next.js operator console that chains LLM planner and worker agents through ScopeSentry-style asset graphs. Microsoft describes it as a system of LLM-driven agents that run an autonomous penetration test from reconnaissance through data exfiltration with little needed human interaction — the same characterization in its Storm-0940 post. Check Point earlier measured thousands of live panels from exposed operator logs in [our Check Point write-up](https://explainx.ai/blog/check-point-hacking-on-autopilot-artex-1608-live-servers-operator-logs-400791-agent-messages-2026); Censys complements that by showing how many consoles answer on the public internet without authentication.\n\nMicrosoft ties ARTEX to OAuth application abuse after the ConnectWise Cloud Backup incident. Even if your company never installed ARTEX, you may still need a service-principal review if a vendor application in your tenant was swept up in Storm-0940 activity — the risk is token reuse, not the presence of the framework on your network.\n\n## Geography and hosting fingerprints\n\nCensys tables show heavy concentration in the United States, Hong Kong, mainland China, Japan, Singapore, and South Korea — not evenly distributed across every economy, but wide enough to invalidate assumptions that this is a single-country problem.\n\nTwo hosting patterns stand out in the write-up. PEG Tech (AS54600) accounts for 493 hosts that share one of two identical port layouts across 13 /24 networks, with 125 hosts each in two of those ranges. Yancy Limited (AS138415) accounts for 157 hosts, mostly inside 156.234.0.0/16, with CyberPanel-style footprints. Censys explicitly compares these to its December 2025 Cobalt Strike listener surge on Yancy, where hundreds of IPs collapsed to a few real servers. Treat raw host counts as an upper bound until your own pivot work separates templates from operators.\n\n*Image: [Censys](https://censys.com/blog/artex-agentic-asbestos/?ref=explainx)*\n\n## How Censys built the inventory\n\nCensys describes matching publicly reachable hosts to the ARTEX operator UI using the page title together with Next.js static-asset fingerprints that ship in the dashboard bundle. That layered signal is narrower than “anything on port 8787,” which matters when you argue with leadership about false-positive rates.\n\nDefault ARTEX often binds **TCP 8787**. The same post’s port table still shows heavy use of **8790**, **443**, and **80** when operators park the panel behind a reverse proxy or move the console onto ordinary web ports. Security teams that only sweep non-standard ports will miss instances where someone terminated TLS on a load balancer and forwarded to a local listener.\n\nCensys anchored its headline count on data through **November 4, 2026**. When you reproduce the hunt internally, snapshot the same window in your ASM or VM tool if you need apples-to-apples trending — day-to-day exposure will drift as operators fix mistakes or rotate IPs.\n\n## Co-located offensive stacks\n\nOne hundred fifty-five hosts ran another Censys-flagged threat alongside ARTEX. The blog calls out concrete overlaps: Cobalt Strike teamservers and Adaptix C2 on the same IP as ARTEX and PentAGI; Vshell and FRP tunnels next to operator panels; CyberStrikeAI co-hosted on 70 hosts, with PentAGI, HexStrike AI and RedAmon also appearing. Censys says operators appear to use several agent frameworks side by side. For other AI-run intrusions, see [our October 2026 cyberattacks tracker](https://explainx.ai/blog/october-2026-cyberattacks-list-breaches-exploits-ai-run-attacks-tracker-2026).\n\nForty-three hosts also ran LLM gateway or API reseller stacks (CLIProxyAPI, New API). That dovetails with CrowdStrike’s South Korean finance narrative about DeepSeek access through resellers — your SOC should alert on new egress to unknown OpenAI-compatible endpoints from DMZ segments, especially when paired with automation coming from the same ASN patterns Censys lists.\n\n*Image: [Censys](https://censys.com/blog/artex-agentic-asbestos/?ref=explainx)*\n\n## Open directories are the silent leak\n\nCensys lists live examples where directories exposed GodPotato, pass-the-hash kits, Linux beacons, HTA droppers, and even operator `.ssh` folders. Because many listings last only hours, defenders without historical scans miss them — threat intel feeds that retain HTTP body history win here.\n\nIf your team already studies agent misuse, compare this with [Google’s LLMjacking report](https://explainx.ai/blog/google-gtig-llmjacking-llm-mixers-1-million-stolen-cloud-ai-access-what-to-do-2026): attackers want cheap GPU plus cheap API keys, not necessarily novel zero-days every week.\n\nCheap mistakes amplify cheap keys. A directory listing that exposes an implant archive does not require a new model capability — it requires one mis-set Nginx flag and a scanner that archives bodies. For teams without Censys-class history, partner with a feed vendor that retains HTTP responses over time or run your own canary crawler against high-risk ASNs your threat model already cares about.\n\nDigital fingerprint motif suggesting exposed identity and credential surfaces on misconfigured offensive security hosts\n\n## How Microsoft frames abused ARTEX\n\nMicrosoft’s October 21 Storm-0940 post is the best primary companion to Censys for readers who need identity remediation language, not just port scans. Microsoft states that Storm-0940 used ARTEX to accelerate the speed of compromised application abuse to authenticate to Microsoft 365 customer environments through applications that had already been granted access — letting the actor operate without interactive user sign-in. In multiple affected environments, Microsoft observed AI-accelerated follow-on activity through Microsoft Graph, including reconnaissance, application credential changes, and attempts to modify privileges that could enable expanded or persistent access.\n\nThe same post gives scale: more than 2,250 tenants saw sign-in attempts in that campaign wave, and multiple environments moved from targeting to privileged access in under three hours when automation succeeded — a reminder that log volume alone will not save you if mean time to contain stays human-paced.\n\nMicrosoft also documents a password-spray phase between October 10 and 14, 2026, from Storm-0940 infrastructure at 144.172.64.99 across eight organizations, with telemetry capturing more than 151,000 failed logon events and peaks near 2,900 failures per minute. That activity did not, in Microsoft’s view, yield successful interactive sign-ins — but it sets up the narrative that the same address space and trust-abuse patterns surfaced again once ConnectWise application identities were in play.\n\nFor incident responders, the mitigation section matters as much as the attribution: keep affected ConnectWise application connections disabled until ConnectWise publishes formal restoration guidance, do not delete enterprise applications as a first response without preserving evidence, inventory the listed application IDs, review every tenant where those apps were ever installed, and preserve Entra sign-in and audit trails rather than waiting for “clean” exports.\n\nThe key line for blue teams is not “new malware” but “abused already trusted applications” — the same connector logic you approved for backup becomes machine-speed lateral movement when credentials spill.\n\nThat story intersects with Censys because both stress operator hygiene: exposed consoles leak TTPs and sometimes API keys; compromised SaaS apps leak tenant trust.\n\n## Where ARTEX came from, and what changed in October\n\nCensys notes ARTEX appeared on GitHub as Autumn-27/ARTEX on July 26, 2026 under the AGPL-3.0 license. Its author also wrote ScopeSentry, a Chinese asset-discovery platform. After CrowdStrike’s October 7 report on a South Korean finance campaign, the author posted a statement saying they were not involved in the attacks and that ARTEX was meant for authorized scopes, and said the project would go closed-source. Censys says the repository was taken down, but archived copies and rebrands such as ScopeWeaver still circulate. TechCrunch adds that Censys’s Silas Cutler stressed that not every exposed instance is malicious, since many belong to authorized testers.\n\n## Red team, lab, and abuse cases\n\nDual-use frameworks are normal in security research. The Censys story is not “ban the repo.” It is **separate authorized labs from production credentials**.\n\nIf your red team runs ARTEX, treat the console like any other C2 panel: no public listener, mutual authentication on the management plane, and no shared API keys with developer OpenAI-compatible gateways used for product features. If your engineers experiment with the same stack in a sandbox, isolate egress so a mistaken route table change cannot spray tokens toward a reseller you do not recognize.\n\nCompliance teams should align this telemetry with vendor questionnaires: any third party that claims “we only run internal scanners” but shows up in ASM with an exposed ARTEX title string has a documentation problem worth a finding, not a shrug.\n\n## What to do Monday morning\n\n1. Scan your attack surface for Next.js ARTEX fingerprints on unexpected ports — Censys publishes a starter query in its post for Censys Platform users.\n2. Force mutual TLS or VPN for any authorized ARTEX instance; if it must face the internet, put it behind SSO with hardware FIDO enforcement.\n3. Rotate LLM keys stored on those hosts — Check Point already showed operators paste tokens in logs in [its operator-log study](https://explainx.ai/blog/check-point-hacking-on-autopilot-artex-1608-live-servers-operator-logs-400791-agent-messages-2026) .\n4. Review OAuth grants and service principals exactly as Microsoft prescribes for Storm-0940 — preserve sign-in and audit data starting at least from October 15, 2026, and extend backward where logs allow.\n5. Add agent guardrails where engineering teams experiment with offensive LLM frameworks. [AgentBeam](https://agentbeam.com/?ref=explainx) , the agent security platform from the explainx.ai team, blocks AI agents before they take dangerous actions.\n\nTeams securing coding agents should pair this with [GitBook’s token-rotation incident write-up](https://explainx.ai/blog/gitbook-security-incident-ai-agents-attacker-api-tokens-revoked-rotate-secrets-2026) — the failure mode is always the same: automated hands plus long-lived credentials.\n\nIllustration of leaked data packets leaving a server stack, representing exposed implants and credentials on misconfigured attack consoles\n\n## Related reading\n\n*Telemetry and quotes follow Censys (November 5, 2026), Microsoft Security (October 21, 2026) and TechCrunch (November 6, 2026). Confirm indicators in your own tools before blocking traffic.*", "url": "https://wpnews.pro/news/censys-1465-internet-exposed-artex-consoles-many-with-open-directories-and-dual", "canonical_source": "https://explainx.ai/blog/censys-artex-1465-exposed-consoles-open-directories-november-2026?utm_source=rss&utm_medium=feed&utm_campaign=blog", "published_at": "2026-11-06 00:00:00+00:00", "updated_at": "2026-11-06 20:23:30.064308+00:00", "lang": "en", "topics": ["ai-agents", "ai-safety", "ai-tools", "artificial-intelligence"], "entities": ["Censys", "Silas Cutler", "ARTEX", "Microsoft", "Storm-0940", "ConnectWise Cloud Backup", "PEG Tech", "Yancy Limited"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/censys-1465-internet-exposed-artex-consoles-many-with-open-directories-and-dual", "markdown": "https://wpnews.pro/news/censys-1465-internet-exposed-artex-consoles-many-with-open-directories-and-dual.md", "text": "https://wpnews.pro/news/censys-1465-internet-exposed-artex-consoles-many-with-open-directories-and-dual.txt", "jsonld": "https://wpnews.pro/news/censys-1465-internet-exposed-artex-consoles-many-with-open-directories-and-dual.jsonld"}}