from prompting to autonomy:
Strategic Leaps in Threat Actors’ Use of AI
Over the past year, Google’s Threat Intelligence Group has tracked a critical shift. As adversaries mature beyond simple generative AI prompts, defenders face a new frontier: fully autonomous, multi-agent offensive frameworks. Aurora Blum leverages frontline telemetry to examine the mechanics of these agentic frameworks and lessons from campaigns in the wild. We will also examine real-world case studies of attacks targeting the enterprise AI stack itself, including "LLMJacking" and compute resource hijacking, to equip threat hunters and strategic defenders with the essential intelligence needed to anticipate and disrupt the next generation of machine-speed, agent-driven offensive tradecraft.