Executives at major cybersecurity firms say emergency breach response is straining capacity, while separate research documents AI agents aiding intrusions across multiple countries, a challenge for companies and responders worldwide.
CrowdStrike said in an Oct. 29 report that VAPOR PANDA began a campaign on Oct. 17 using the ARTEX agentic framework. The operation ran for 11 days, with targets spanning at least 14 countries; the number of organizations targeted exceeded 200, the company said. CrowdStrike confirmed follow-on activity at nine organizations. It said the operation used the Chinese-developed DeepSeek large language model and a second subscription-based commercial large language model it could not confirm. The report said a human operator oversaw agents working across multiple targets at once. CrowdStrike described the activity as China-nexus but did not establish a link to Chinese intelligence.
In incident cases dating from late August 2026, Google Threat Intelligence Group and Mandiant identified attackers using agentic tools including ARTEX, PentAGI and CAIRN. The agents handled tasks such as moving through networks, exploiting databases and preparing data for theft. CrowdStrike separately identified four actors whose AI agents selected actions and ran commands in campaigns between May and August 2026. Google and Mandiant compared the use of agentic frameworks with earlier cases in which attackers repurposed penetration-testing tools such as Cobalt Strike and Metasploit.
The capacity concerns come from separate reporting. NBC News reported Monday, Nov. 9, that executives at Mandiant, CrowdStrike and Kroll said some organizations seeking emergency incident-response help had been turned away. NBC gave no count. Mandiant Chief Technology Officer Charles Carmakal told NBC that demand had accelerated in the two months before the report, after a busy year.
Kroll risk advisory president Brent Tomlinson told City AM Wednesday, Nov. 11, that top incident-response providers had limited room for new cases. Marsh, an insurance broker, said Thursday, Nov. 12, that incident volumes were growing faster than firms could hire and train responders. Marsh said retainer clients may receive priority, but being on an insurer’s response panel does not guarantee reserved capacity. Organizations without retainers may receive help on a best-efforts basis, depending on provider capacity.
The reports do not show that AI explains the broader incident-response workload. Google’s 2026 M-Trends report, based on Mandiant investigations from 2025, said attackers used AI for reconnaissance, social engineering and malware development. It also said human mistakes and weaknesses in organizational systems remained at the root of most successful intrusions.
