An operational assessment of how cybercriminal and state-sponsored threat actors use generative AI across the attack lifecycle, from malware and exploit development to social engineering, reconnaissance and hands-on-keyboard operations. It draws on public disclosures and QuoIntelligence's own observations from December 2024 to May 2026, and finds that AI mostly lowers the cost of running existing attack patterns rather than creating new ones. Written for security leaders and defenders, it closes with defender priorities for 2026.