Critical vulnerabilities have been discovered in major AI agent frameworks used by enterprises to build applications, highlighting a security failure that goes beyond prompt injection. Researchers from Check Point have identified nearly a dozen flaws, some of which are critical, in these frameworks, allowing prompt-controlled content to breach the boundary into trusted framework logic1. This means that a bug in an agent framework can have far-reaching consequences, compromising the entire system. The discovery underscores the need for a more comprehensive approach to securing AI systems, one that extends beyond individual models or prompt injection vulnerabilities. The vulnerabilities in question affect the underlying frameworks that power AI applications, making them a more significant concern for enterprises. This matters to practitioners because it emphasizes the importance of securing AI agent frameworks to prevent potential exploits and data breaches.