CrowdStrike says the China-nexus group VAPOR PANDA used ARTEX, an agentic AI framework, to run parallel intrusions against more than 200 organisations from 17 October 2026, including semiconductor, pharmaceutical, nuclear and defence organisations in the US. No operational technology impact has been reported, but the techniques used (stale keys, leaked VPN configurations, unsigned SAML and exposed services) are exactly the paths that lead from corporate IT into industrial networks. OT operators should rotate old credentials, close weak remote access routes and confirm that the IT/OT boundary would hold if an AI-driven intruder reached the enterprise domain.
What happened
In a blog post dated 29 October 2026, CrowdStrike reported that its OverWatch team observed a single ARTEX controller, which CrowdStrike Intelligence links to VAPOR PANDA infrastructure, directing exploitation activity against entities in at least 14 countries over an 11-day window. Industrial Cyber covered the research on 5 November 2026.
According to CrowdStrike, the campaign ran mainly on the DeepSeek large language model plus a second commercial model it could not confirm. ARTEX coordinates a main agent with planner and worker agents that share a graph of facts, findings and goals. A human operator stayed in the loop, steering the agents at key decision points, while the agents worked through reconnaissance, credential theft, web shell deployment and lateral movement across several targets at once.
Targets named by CrowdStrike include foreign ministries in Latin America, Europe, Asia and the Middle East, telecommunications providers in the US, Belgium and the Philippines, and a separate engagement against 19 US organisations in the pharmaceutical, semiconductor, defence, nuclear and financial sectors. CrowdStrike says it confirmed post-exploitation activity at nine organisations, where the group deployed the BLOODALCHEMY backdoor and carried out credential spraying, network scanning and reconnaissance.
The examples CrowdStrike describes are instructive because none of them is exotic:
- At one US pharmaceutical organisation, the agent chained an unsigned SAML assertion with an XML external entity (XXE) flaw to reach the cloud instance metadata service, gaining a role that could issue commands to thousands of instances.
- At another, it abused an unauthenticated orchestration service to gain root in a production container environment and read LAPS passwords.
- Against a Greek government entity, the operator fed the agent a 2022 mail server directory export containing keys that had likely never been rotated, then used VPN configurations found in a user's sent mail to reach the internal network.
- The operator also told the agents to find Windows hosts that lacked an endpoint sensor, so follow-on activity could focus on unprotected machines.
CrowdStrike assesses that this activity is separate from ARTEX use reported by industry sources, including Microsoft, under the Storm-0940 name, and that China-nexus groups will highly likely keep adopting agentic frameworks over the next 12 months.
Why it matters for OT
VAPOR PANDA has, according to CrowdStrike, historically targeted government, energy and extractive organisations to build long-term access. This campaign broadened that to semiconductor, pharmaceutical and nuclear organisations: sectors where production, quality and safety depend on industrial control systems.
The reported intrusions stayed in IT, cloud and identity systems. That is still relevant to OT for three reasons.
First, many OT incidents begin in IT. Shared Active Directory domains, jump hosts reachable from the corporate network, and engineering laptops that move between zones are common in manufacturing and process plants. An attacker holding domain administrator rights may be only one weak conduit away from the control network.
Second, the entry points are old hygiene problems. Unrotated keys sitting in exports and backups, VPN profiles shared by email, and unauthenticated management services are frequently found around OT remote access, vendor support tunnels and historian or MES servers that bridge IT and OT.
Third, agentic tooling changes the economics. CrowdStrike notes that it lets a small team try many attack paths in parallel and switch quickly when one fails. A flat network or an inconsistent control that a human attacker might never get round to testing becomes cheaper to find. For OT, where patching is slow and legacy systems are common, defence has to rely on segmentation, strong identity and good visibility rather than on speed of patching alone.
Who should act
- Semiconductor, pharmaceutical, defence and nuclear-related manufacturers, given the sectors named in CrowdStrike's report.
- Energy, utilities and extractive industries, which CrowdStrike lists among VAPOR PANDA's historical targets.
- Any industrial organisation whose OT relies on corporate identity services, shared VPN infrastructure or cloud-hosted engineering and data platforms.
In the EU, organisations in scope of NIS2 are expected to manage access control, supply chain and incident handling risk, and a campaign like this is a useful test case for those measures. In the UK, operators working to the NCSC Cyber Assessment Framework can use it to check identity and remote access outcomes. For guidance by region, see our standards overview.
What to do now
- Rotate long-lived secrets. Following CrowdStrike's advice, rotate signing keys, pre-authentication keys and service account credentials after any suspected exposure, including those held in old backups and directory exports. Include OT service accounts, historian and SCADA integration accounts, and shared vendor accounts.
- Audit OT remote access. List every route into the OT network: VPNs, jump hosts, vendor tunnels and remote desktop tools. Remove VPN profiles and keys that have been shared by email, require multi-factor authentication, and put access behind an approval and time-limited process. See remote access control.
- Separate OT identity from the enterprise domain. Where OT systems trust the corporate Active Directory, plan a separate OT domain or tightly controlled trust, so a compromised enterprise administrator account does not become an OT administrator account.
- Check the IT/OT conduits. Confirm that firewall rules between the enterprise network, the industrial DMZ and control zones allow only the flows that are documented and needed. Our guide to zones and conduits explains the approach.
- Close exposed management services. Look for unauthenticated orchestration, container or web management interfaces on servers that sit near OT, such as historians, MES and patch servers, and patch internet-facing appliances as CrowdStrike recommends.
- Find the blind spots. Attackers in this campaign looked for hosts without endpoint protection. Know which OT-adjacent Windows hosts have EDR or OT-compatible monitoring and which do not, and use network monitoring where agents cannot be installed safely.
- Hunt for the published indicators. CrowdStrike's report lists ARTEX markers and BLOODALCHEMY indicators. Ask your security operations team to search IT and IT/OT boundary systems for them.
How OTVista helps
If you want support, OTVista researchers can help. Related services: OT security architecture review. The free OT Security Quick Check is a quick way to see where to start.
Sources: CrowdStrike, VAPOR PANDA Uses AI-Driven ARTEX to Scale a Global Espionage Campaign; Industrial Cyber, VAPOR PANDA hackers use multi-agent AI to target government, pharmaceutical, semiconductor, defense organizations
Topics: VAPOR PANDA, ARTEX agentic AI, AI-driven cyberattacks OT, IT/OT boundary security, OT remote access security, industrial cybersecurity threats
