8 insights from Proofpoint Protect: Security bets on intent as AI agents join the workforce
SiliconANGLE Cheryl Knight
Proofpoint says AI agents are now a security problem, not just a productivity tool. The fight is shifting to intent: what humans and agents mean to do, not just what they touch.
Based on reporting by SiliconANGLE, Cheryl Knight — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
At Proofpoint Protect in San Diego, the mood around AI agents had changed. The question was no longer whether enterprises would use them. It was how they would keep them from wandering off script once they had access to inboxes, data and systems.
That shift explains why the word “intent” came up so often. Proofpoint is building around knowledge graphs and models that try to judge whether a person or an agent is actually doing what it is supposed to do. That matters because an agent given a job can optimize itself into something else, and attackers are already using AI to crank out polished phishing and string exploits together at machine speed.
Proofpoint also rolled out two new agentic systems for collaboration security and data and AI security, both tied to a knowledge graph that maps how people and AI communicate and reach data. The company says it cannot patch every vulnerability, so it wants compensating controls that can turn written policy into live enforcement.
The endpoint side is becoming just as important. Proofpoint said about 99% of the agentic activity it monitors runs on ordinary endpoints, not in the cloud, which puts its sensor network at the center of agent security. And as AI agents start acting more like insiders, governance has to become real-time, not a document sitting in a folder no one opens.
Anthropic is pushing the same basic warning from the other side of the table. It held back general release of its Mythos Preview model to build guardrails and get it into defenders’ hands first, after noting that new models can chain low-severity flaws into more serious exploits. Its advice is simple enough: write policy, sandbox, watch closely, and start small so the blast radius stays visible.
The business side is moving fast too. Proofpoint said annual recurring revenue is close to $2.5 billion and growing nearly 20%, including Hornetsecurity, after doubling since Thoma Bravo took it private in 2021. Customers are consolidating vendors to make room for AI spend, which tells you everything about where the budget pressure is coming from.
My take — AI-written commentary, not fact-checked reporting
The smart money is finally treating AI agents like junior staff with sharp knives, not magic software. That is long overdue. The industry’s favorite habit is to ship first and write the guardrails later, which works right up until an agent has inbox access and a loose idea of “helpful.”
Read more about this at: SiliconANGLE