CrowdStrike Says China-Linked Hackers Aimed DeepSeek-Run AI Agents at 200+ Organizations in 11 Days
CrowdStrike says a China-linked group used an AI agent framework running mostly on DeepSeek to target more than 200 organizations in 11 days, with nine confirmed intrusions. The scale is the story, and the attribution is the company's own.
One operator. Two hundred targets. Eleven days. That is the pitch CrowdStrike is making about a China-linked hacking group it calls VAPOR PANDA, and it is a bleak one for anyone who still thinks AI in cyber offense mostly means better phishing emails.
In a report published October 29, CrowdStrike says its OverWatch team found VAPOR PANDA running an AI agent framework called ARTEX alongside DeepSeek, the Chinese-developed large language model, plus a second commercial model the company could not identify. From October 17, CrowdStrike says, the group pointed ARTEX at more than 200 distinct organizations. Activity from a single controller touched entities in at least 14 countries over an 11-day window.
The list reads like an intelligence agency wish list. CrowdStrike names foreign ministries in Latin America, Europe, Asia and the Middle East, plus semiconductor, defense and think tank organizations in the U.S. and telecom providers in several regions. One engagement alone went after 19 U.S. organizations in pharmaceuticals, semiconductors, defense, nuclear and finance.
Now the part that deserves some skepticism. Two hundred is organizations targeted, not breached. CrowdStrike says it confirmed post-exploitation activity at nine of them, where the attackers got in, deployed a backdoor called BLOODALCHEMY, and moved on to credential spraying, network scanning and reconnaissance. It also saw password spraying against at least 12 more, including U.S. think tanks and European government bodies. Nine confirmed intrusions is not a rounding error for a ministry of foreign affairs.
The attribution, for what it is worth, is CrowdStrike's own. The company calls the actor China-nexus. NBC News describes the hackers as Chinese state-linked, citing CrowdStrike.
"This campaign marks an escalation in China-nexus use of AI agents, from single intrusions to multiple simultaneous engagements directed by a human operator." CrowdStrike
That last phrase matters. This was not a machine running loose. CrowdStrike says recovered transcripts show an operator issuing instructions in Chinese, tagged with a label meaning "human instruction" and timestamped in China Standard Time. The agents did the grinding, and the human stepped in at decision points to steer them. The controller intermittently left its tooling, target files and agent transcripts exposed, which is how researchers got a look inside.
The scale signal is the token count. In a single intrusion, CrowdStrike says VAPOR PANDA's planning agents consumed on the order of 1.4 billion input tokens, which the company says implies far higher consumption across the whole operation. CrowdStrike also reports the operator telling agents to find Windows machines in the domain without the Falcon sensor, which is the company's own product. A security vendor reporting that attackers tried to dodge its software is not a neutral data point, so read accordingly.
On DeepSeek, be precise. CrowdStrike says the group ran ARTEX primarily on the model. The report describes DeepSeek as the tool used, not as a collaborator, and it suggests the mix of providers was a deliberate attempt to build a resilient toolchain rather than depend on a single service.
ARTEX itself is not new on this site. We covered it last week in our report on Microsoft's ConnectWise findings. CrowdStrike is explicit that the VAPOR PANDA campaign is separate from the ARTEX activity that industry sources reported under the Storm-0940 name, and says that likely shows the framework is spreading among several China-nexus groups. That is CrowdStrike's assessment, and I am not treating the two actors as the same.
CrowdStrike also says this is VAPOR PANDA's first observed use of AI-driven intrusion tooling. Since 2021, it says, the group has mostly gone after government, think tank, energy and extractive organizations to build long-term access. Agents let a small team widen that net into pharma, chips and nuclear without hiring a hundred more operators.
The company expects China-nexus actors will "highly likely" keep adopting frameworks like ARTEX over the next 12 months. That is a forecast, and a vendor's one at that. The campaign it describes, if CrowdStrike has it right, is already enough to worry about. Each extra target costs the attacker little, while each defender still has to be right every time.
