As the use of AI agents rises within enterprises, so do the cybersecurity risks of deploying them without continuously verifying that they follow the intent and governance set by corporate users.
To address this, cybersecurity and compliance vendor
Proofpoint recently introduced Proofpoint AI Security, a SaaS platform that delivers intent-based verification across endpoints, browsers, Model Context Protocol (MCP) agent connections, and other environments where AI is used.
Also introduced is Proofpoint’s new Agent Integrity Framework, which outlines how autonomous AI agents should be designed with built-in operational integrity. It includes a five-phase maturity model for implementing AI agents in business workflows, from discovery through runtime enforcement. Agent integrity is intended to ensure that AI agents operate within defined boundaries, including authorized permissions and expected behaviors throughout development and use.
For enterprises using AI agents, the new tools provide insights into the semantic content of AI interactions, where risk often resides. Traditional cybersecurity tools typically cannot assess this layer. Using Proofpoint AI Security, enterprises can observe whether an AI agent’s actions align with the intent and context of assigned tasks.
This is achieved through intent-based detection models that compare original requests, defined policies, and intended purposes with the observed behavior of AI agents. The platform analyzes the full semantic context of AI interactions and flags misaligned or high-risk actions in real time, before damage occurs.
Protecting Against AI Agent Attacks
AI agents deployed in enterprises face growing risks from cybercriminals using their own autonomous agents to manipulate sanctioned systems into unintended behaviors. These attacks can create new entry points within an organization’s defenses. Addressing this threat is a core function of Proofpoint’s new platform.
Tim Choi, group vice president for product go-to-market and strategy for AI security at Proofpoint, said the new tools are designed to help the company’s channel partners tackle these emerging risks for enterprise customers.
“Enterprises are rapidly deploying generative AI and autonomous agents, and both customers and partners are telling us they are concerned about governance gaps, data exposure, and the expanding attack surface created by AI-driven workflows,” said Choi.
The Proofpoint AI Security platform will initially be delivered through value-added resellers, systems integrators, and resellers.
“At this stage of the market, we are focused on enterprise deployments, where large organizations want direct visibility and control over how AI and agents operate within their environments,” Choi said. “For that reason, Proofpoint AI Security is currently being made available through our enterprise channel partners rather than managed service providers (MSPs). As AI adoption matures, particularly within the SMB segment, we will evaluate expanding availability through MSPs.”
What’s Under the Hood
Designed to bring visibility and control over AI agents used within enterprises
, Proofpoint AI Security scours a company's infrastructure and developer environments where AI is being used, including endpoints, browser extensions, and MCP connections to determine where there are AI risks and what needs to be fixed. These evaluations are especially important where developers are using agent-connected coding assistants, plugins, and MCP-integrated tools to do their work, according to Proofpoint.
Using Proofpoint AI Security, enterprises can discover which sanctioned and unsanctioned AI tools are being used by humans or AI agents, including OpenClaw, Ollama, ChatGPT, and MCP, and then enable rules and policies to allow or restrict them. The SaaS platform can also observe prompts, responses, and data flows during AI tool usage, while also providing access controls and guardrails on AI usage. In addition, runtime inspections and policy enforcements can also be enabled during live AI interactions to observe their behaviors.
As organizations rapidly deploy autonomous AI agents to browse the web, access internal systems, send emails, execute code, and orchestrate workflows, risks such as agentic privilege escalation and zero-click prompt injection attacks are becoming real-world threats. A single AI request can trigger dozens of autonomous actions across multiple systems, often at machine speed and without human oversight. Existing security tools can see traffic and permissions, but they don’t evaluate whether AI behavior aligns with the original user intent.
