Depth vs breadth: the two kinds of AI agent security testing

Most agent security tooling focuses on finding new vulnerabilities. But who's checking whether last month's prompt injection still works after a model