Vapor Panda Uses Agentic AI to Breach 140 Sensitive Targets in 11 Days

CrowdStrike reports that Vapor Panda, a China state-backed threat group, used ARTEX, an automated agentic AI penetration-testing framework, paired with DeepSeek to conduct a fast-moving campaign against more than 200 organizations across at least 14 countries. Since October 17, the group compromised about 140 sensitive targets in 11 days, according to the threat report described by CyberScoop. CrowdStrike told CyberScoop that about 70% of Vapor Panda’s targets were confirmed compromised or showed evidence indicating moderate confidence of a successful intrusion. The campaign remains active, and the reported scale and tempo make it a relevant threat signal for contractors supporting government missions and handling sensitive information.

The reported targets span foreign ministries, a European country’s presidency and prime minister’s office, defense and intelligence agencies, think tanks, and government organizations across defense, justice, finance, and diplomacy. CrowdStrike also identified 19 U.S.-based organizations in the pharmaceutical, semiconductor, defense, nuclear, and financial sectors, as well as telecommunications providers in the United States, Belgium, and the Philippines. Adam Meyers, CrowdStrike’s senior vice president of counter adversary operations, characterized the campaign as a “diverse operational mission” focused on sectors and geographies that Beijing considers economically or politically important. The report does not provide names of specific government victims, identify a procurement requirement, or point to a contract vehicle or funding source.

The operational pace is central to the procurement implications. ARTEX and DeepSeek reportedly enabled requests as often as every 10 seconds and allowed the group to pursue multiple attack paths concurrently. Meyers contrasted this with hands-on-keyboard activity CrowdStrike had tracked over 15 years, which typically involved up to 40 commands per hour; with ARTEX, he said, activity could exceed 1,500 commands per hour. He also warned, “If you diversify that across 200 targets, it costs them very little to run a mass-exploitation campaign and drop a websell.” For contractors, this points to the importance of security capabilities that can identify and contain coordinated activity across endpoints, identities, and connected domains—not simply detect isolated indicators or rely on investigation processes built for slower campaigns.

This is a cybersecurity demand signal, not a specific procurement opportunity. No agency requirement, solicitation, award, dollar amount, procurement contact, or funding source is identified. Contractors can nevertheless use the campaign pattern to inform internal security capability reviews and future market research: assess whether monitoring can handle high-frequency activity, whether identity controls can surface abnormal access, and whether response teams can correlate signals across environments quickly enough to contain parallel attack paths. CrowdStrike described agentic adversary activity as growing month over month at almost threefold, reporting roughly 45 such attacks in September and 126 in October at the time of the report, with the October figure still climbing. The source distinguishes this active Vapor Panda campaign from separate ARTEX activity attributed by Microsoft Threat Intelligence to Storm-0940 earlier in the month, including a supply-chain attack targeting ConnectWise customers.

• Threat actor and tool: CrowdStrike tracks the China-backed group as Vapor Panda; the campaign used the ARTEX agentic AI penetration-testing framework with DeepSeek.
• Campaign scale: More than 200 organizations in at least 14 countries were targeted; about 140 sensitive targets were compromised in 11 days.
• Reported success indicators: CrowdStrike said about 70% of targets were confirmed compromised or had evidence suggesting moderate confidence of a successful intrusion.
• Operational speed: Requests reportedly occurred as often as every 10 seconds; ARTEX activity could exceed 1,500 commands per hour, compared with up to 40 commands per hour in the historical hands-on-keyboard activity described by Meyers.
• Sectors and entities: Targets included foreign ministries, defense and intelligence agencies, a European presidency and prime minister’s office, think tanks, and 19 U.S.-based organizations in pharmaceutical, semiconductor, defense, nuclear, and financial sectors.
• Telecommunications exposure: Providers in the United States, Belgium, and the Philippines were among the reported targets.
• Contractor action: Review endpoint and identity monitoring, cross-domain threat visibility, and incident-response capacity for fast, coordinated activity spanning multiple attack paths.
• Market classification: This is a cybersecurity demand signal, not an identified solicitation or award; the report names no agency requirement, contract vehicle, procurement contact, or funding source.
• Campaign distinction: The active Vapor Panda activity is separate from ARTEX activity that Microsoft Threat Intelligence attributed to Storm-0940, including a reported supply-chain attack targeting ConnectWise customers.
• Trend indicator: CrowdStrike observed roughly 45 agentic adversary attacks in September and 126 in October, with activity described as climbing.