Box has announced new security and governance capabilities designed to give organizations greater control over AI agents accessing and acting on enterprise content. The features will apply to Box-native agents and third-party systems such as Claude, ChatGPT and Gemini.
The company is extending its existing content security framework to workflows in which AI agents search, analyze, create, modify or share files. Box said the controls are intended to help businesses deploy agents at scale without allowing them to access content beyond the scope of an assigned task.
Box cited research showing that security, regulatory and trust concerns remain major barriers to enterprise AI adoption. Its 2026 State of Enterprise AI report found that 90% of surveyed IT leaders viewed these issues as the largest obstacle to giving AI agents access to company content.
The new capabilities operate within the Box platform, where an organization’s documents and other business information are already stored. This content-level approach is designed to make agent actions permissioned, visible and auditable without requiring customers to deploy a separate security product.
Agent guardrails will allow administrators to define what custom Box AI agents can and cannot do based on company policies and the sensitivity of the content involved. Organizations can apply label-based access restrictions, require approval before an agent deletes content and prevent external sharing.
Box is also adding prompt-injection detection to inspect inputs before they reach an AI model. Administrators will be able to log, flag, or block suspected attempts to manipulate an agent into disregarding its instructions or revealing protected information.
Controls for Box’s Model Context Protocol server will govern the actions of connected external agents. Administrators could permit file creation only in approved folders, block external sharing or limit content moves to designated locations.
Classification-based policies will allow organizations to exclude designated content from agent searches and access. These restrictions can apply to both third-party agents and custom AI agents built within Box.
Agent activity oversight will provide visibility into how external systems interact with customer content. Companies can establish threshold-based alerts to identify unusual behavior and respond to potentially unauthorized activity.
Box will maintain audit trails with contextual records of each agent session. The company said these records can support compliance requirements, retention policies, legal holds and internal investigations.
Human-in-the-loop controls will require a person to approve certain sensitive or high-impact actions before an agent completes them. This is intended to preserve human oversight when an automated action could delete, expose or materially alter important content.
The features could be used by financial services firms to protect transaction materials, healthcare organizations to secure patient information and law firms to govern contract and discovery workflows. Insurance companies could also use the controls to restrict access to policyholder and claims records.
Box plans to make the new security and governance features available to customers using its Enterprise Advanced plan over the coming months.
KEY QUOTES:
“83 percent of organizations are already experimenting with AI agents across their most critical tasks. As these agentic workflows become more deeply embedded in the enterprise, it’s critical to create the proper security controls to ensure agents have what they need to function effectively, without accessing, modifying, or exposing content beyond the scope of its intended task.”
“Box already provides a safe environment for organizations to apply AI to their most critical business knowledge, and with these new controls, we are creating a standard for deploying agents of any kind securely and at scale.”
Manoj Asnani, Vice President of AI Security, Privacy, Compliance and Governance Products at Box
“As we rapidly advance our utilization of AI agents, we expect Box, which has consistently led the development of security management capabilities for secure collaboration, to provide the administrative features needed to safely leverage this new era of AI.”
“In particular, we’ve found it extremely reassuring that Box offers multi-vendor support, allowing us to flexibly switch between AI models, while providing security management capabilities that span prevention, detection, and response. With a protective layer that appropriately manages AI agent access to content, we’re confident our critical content will remain protected as we expand our use of AI.”
Tatsutoshi Murata, Head of the IT Strategy Department at Nomura Research Institute
“As organizations rapidly adopt agentic AI, securing the content layer becomes the critical foundation for deployment. Box’s new security and governance controls address the primary barriers of privacy and unauthorized access directly where the data lives.”
“By embedding guardrails, prompt injection detection, and human-in-the-loop oversight into the platform, Box is establishing a vital trust standard that allows enterprises to confidently scale both native and third-party AI agents across their most sensitive content.”
Amy Machado, Senior Research Director of Content and Knowledge Management Strategies at IDC
