Palo Alto Networks pitches Authority-Aware DLP for AI agents

The article argues that AI security problems have shifted from employees pasting data into ChatGPT to sanctioned AI apps being used with personal accounts, and further to AI agents operating through MCP and A2A protocols. This evolution, it says, forces a fundamental strategy change because data exposure no longer happens only through front-door prompts but across three operational layers.

Palo Alto Networks frames the core issue as traditional security inspecting the payload while missing context, meaning it cannot tell what an agent is trying to do or where data lands next. Their proposed fix is Authority-Aware DLP, which extends Zero Trust principles from humans to agents by understanding agent identities, limiting access, and enforcing policy inline.

The company positions this as necessary because AI agents will not wait for manual steps like copying files or approving transfers. The article suggests approval at one layer should not imply trust at the next, and it ends with an offer for a security audit of local agent footprints and hidden MCP traffic.