At this week’s Proofpoint Protect event, the security debate moved beyond whether enterprises will deploy AI agents to how they will govern them. Attackers now use AI to write flawless phishing lures and chain exploits at machine speed. Meanwhile, the agents companies run internally act as insiders, with access to data, systems and inboxes.

The next phase will be defined by intent. Defenders are building knowledge graphs and intent-based models that judge whether a human or an agent is doing what it should. They are also turning written governance policies into real-time controls. Frontier labs have added urgency, with Anthropic PBC expanding Project Glasswing to put its vulnerability-hunting Mythos Preview model in more defenders’ hands.

“The large enterprises that are competing, they look at the transformation edge that they’re trying to get as the competitive edge. So their game is speed,” said John Furrier, executive analyst for theCUBE Research. “They know that if they can beat the competition with better products and better outcomes, they win. Now, the risk is the security piece.”

Executives and researchers from Proofpoint Inc. and Anthropic spoke to Furrier during Proofpoint Protect in San Diego, with exclusive coverage from theCUBE + NYSE Wired. The interviews covered agentic threats, knowledge graphs, platform consolidation and why this is both the best and worst of times for AI agents.

Here are eight standout insights from the event:

1. Security teams shift from blocking AI to enabling safer adoption.

Customers are approaching agentic AI with cautious optimism, and security teams are being pushed to stop blocking projects and start enabling safe AI use. Identifying the intent of an agent or a human is now the most critical step, because an agent given a job can optimize its way into doing something else entirely, explained Molly McLain Sterling (pictured), senior director of cybersecurity strategy at Proofpoint.

Check out theCUBE’s complete interview.

2. Intent becomes the foundation for two new agentic security systems.

Proofpoint introduced two new agentic systems for collaboration security and data and AI security, both built on a knowledge graph that tracks how people and AI communicate and access data. Because no company can patch every vulnerability, enterprises need compensating controls that judge the intent of every human and agent, according to Sumit Dhawan, chief executive officer of Proofpoint.

Watch the full interview from theCUBE.

3. Tiered intent models target attacks that appear legitimate.

Attackers increasingly hijack legitimate supplier email threads without using malware, so Proofpoint has added intent-based models to its Nexus detection suite that run in Flash, Extended-Thinking and Deep-Thinking tiers. A new Community Hyperloop spreads each new detection to every customer at machine speed, noted Tom Corn, executive vice president and general manager of the Threat Protection Group at Proofpoint.

See theCUBE’s full coverage.

4. AI agents emerge as insider risks concentrated on enterprise endpoints.

About 99% of the agentic activity Proofpoint monitors runs on ordinary endpoints rather than in the cloud, which makes its tens of millions of endpoint sensors the foundation for agent security. Governance policies written for humans must be turned into real-time controls that agents can understand, emphasized Ryan Kalember, chief strategy officer of Proofpoint.

Don’t miss the full interview on theCUBE.

5. Anthropic urges enterprises to engineer for trust before scaling agents.

Anthropic held back general release of its Mythos Preview model to build guardrails and get it to defenders first, since new models can chain low-severity vulnerabilities into serious exploits. Companies should write policy, set up sandboxing and visibility, and start small with a clear view of the blast radius, advised Robert Bair, head of national security partnerships at Anthropic.

Hear the full story on theCUBE.

6. Proofpoint nears $2.5B in annual recurring revenue as customers consolidate.

Proofpoint’s annual recurring revenue is close to $2.5 billion and growing nearly 20%, including Hornetsecurity, after doubling since Thoma Bravo LP took it private in 2021. Security chiefs are consolidating providers to free up budget for AI, including a Canadian bank that replaced four suppliers in a $25 million, five-year deal, shared Remi Thomas, chief financial officer of Proofpoint.

Catch the complete conversation on theCUBE.

7. AI divides cyberattackers into increasingly skilled and sloppy camps.

Skilled threat actors use AI to speed up malware development and copy lures into as many as 16 languages, while low-end actors are getting sloppier. Static detections now lose their value quickly, so defenders must build dynamic ones that anticipate attackers’ next moves, highlighted Selena Larson, principal threat researcher at Proofpoint.

Watch theCUBE’s full sit-down.

8. Knowledge graphs become the control layer for human and AI access.

As humans and AI agents increasingly access data directly, enterprises need a way to understand both why access is happening and what each actor is allowed to touch. Proofpoint is building around a knowledge graph that connects actors, actions and data, creating a contextual layer for governance, detection and remediation as AI systems operate continuously, according to Mayank “MC” Choudhary, EVP and GM of the Data Security and Governance Group at Proofpoint.

Check out the complete discussion.

Stay tuned for the complete Proofpoint Protect interview playlist.

Photo: SiliconANGLE