ESET has released its 2026 H1 Threat Report. In addition to the misuse of more than 3,000 AI agent skills, the report describes new techniques including PromptSpy, an Android malware that uses generative AI at runtime.
Key threat trends in the first half of 2026 included the identification of more than 3,000 malicious AI agent skills as agentic AI became more widespread, as well as the discovery of PromptSpy, the first Android malware to use generative AI at runtime. Concerns have been raised about agent skills in connection with supply chain risks involving external libraries and additional skills, highlighting the need to establish AI governance.
Quishing, or phishing that exploits QR codes, reached an all-time high as attackers targeted gaps in everyday routines, with an average of 100,000 detections per month. Attacks disguised as payslips and employee benefits notices were particularly prominent.
ClickFix, which recorded the highest number of detections in Japan worldwide, saw a sharp increase in social engineering attacks using fake CAPTCHA verification pages. Activity targeting sign-ins to cloud services is intensifying, with Japan accounting for 14% of detections overall.
Other notable trends included a decline in the rate of ransomware ransom payments even as attacks increased, the continued misuse of EDR killers, and intensifying conflicts among cybercrime groups.