Researchers at Palo Alto Networks' Unit 42 have uncovered a series of AI-assisted cyberattacks targeting at least 19 organizations between August and September 2026. The investigation, published on November 12, revealed that hackers gained unauthorized access or retrieved data from 10 of these organizations. The attacks utilized a customized version of Cairn, an open-source framework for AI-driven penetration testing, combined with Zhipu GLM and DeepSeek models.
The operations ranged from brief 20-minute intrusions to prolonged efforts lasting over 50 hours. Human operators selected the targets and maintained control over the AI agents, which executed the attacks. Among the affected entities were semiconductor companies, defense and policy organizations, and civil-society groups. Notably, attackers accessed a semiconductor company's production system and obtained over one million member records from a US teachers' union. They also modified a live standards-ballot document and acquired nonpublic data from an industry standards organization.
Unit 42 assessed with moderate confidence that the activity was linked to China-nexus intelligence collection, based on language, tooling, and target selection. The report emphasized that familiar security weaknesses enabled the campaign, highlighting the importance of identity controls, reducing exposed attack surfaces, and regular software patching as central defenses. The researchers concluded that AI increased the scale and reach of the campaign but not its technical sophistication.