The last year saw generative AI go from a big research advance to something that directly impacts real people and products. The multidimensional interplay between gen AI and security means we must protect AI-powered workloads against old and new risks, and apply gen AI to solve security problems. The resulting collisions pose new challenges to overcome: 

People expect AI to deliver on the initial hype with enduring value. Organizations and users expect AI-powered assistance to be more than a simple productivity boost; it needs to consistently provide correct answers and insights, and maintain context to actually get smarter over time. 

Foundation Models aren’t enough on their own to solve many real-world security problems. Organizations and practitioners have realized that tackling most tasks requires using multiple—often specialized—models, planning, orchestration, extensions, external memory, and other techniques together. Some problems may require open-ended, long-running agents that can run indefinitely to advance specific goals. 

Google Cloud’s Product Vision for AI-Powered Security