We Need a Shared Responsibility Model for AI

Researchers found vulnerabilities across AI tools over several months that could enable data exfiltration, exploitation of AI browsers, and poisoning of AI memory; vendors often dismissed these issues as out-of-scope. The report argues for a shared responsibility model—analogous to cloud "security of the cloud" vs "in the cloud"—to clarify vendor and customer roles across AI-as-SaaS, AI-as-PaaS, and AI-infrastructure layers to prevent widening security gaps.