Box, Inc. announced new security capabilities designed to give organizations greater control over AI agents working with enterprise content. The new features, including agent guardrails, prompt injection detection, and classification-based access policies, extend Box's enterprise-grade security controls to both Box Agents and third-party agents such as Claude, ChatGPT, and Gemini. These capabilities aim to address the primary barriers to scaling AI, as 90% of IT leaders surveyed identified security, regulatory, and trust concerns as the biggest obstacle to granting AI agents access to enterprise content.

Addressing Enterprise Security Challenges

Box's 2026 State of Enterprise AI report highlights that security and privacy are the leading obstacles to deploying AI agents at scale. To mitigate these risks, Box has integrated protections directly at the content layer. This ensures that every agent action is intentional, permissioned, and auditable, allowing organizations to move from limited pilots to production-scale AI deployments without compromising governance. The controls are built to manage agents whether they are built in Box or connected through third-party platforms.

Key Security and Governance Capabilities

The new security and governance features do not require additional tools to deploy. They include several specific controls designed to manage agent behavior and data access:

Industry Adoption and Strategic Vision

The release of these controls aligns with a broader industry shift toward agentic AI. Manoj Asnani, VP of AI Security, Privacy, Compliance & Governance Products at Box, noted that 83% of organizations are already experimenting with AI agents. The new capabilities are intended to create a standard for deploying agents securely by ensuring they access only the data necessary for their specific tasks.

This focus on security complements the evolving technical landscape of AI deployment. Box CEO Aaron Levie has previously emphasized the efficiency of multi-model agent systems, where a frontier model acts as a planner and orchestrator alongside cheaper workhorse models. This strategy, which can result in significant cost improvements by reducing total token usage, relies on routing different models based on the stage of the task. By securing the content layer, Box's new controls provide the foundation necessary for enterprises to confidently adopt these complex, multi-model workflows.

Availability

Box's new security and governance capabilities for AI agents will be rolling out to customers on the E-Advanced plan in the coming months.

Box Inc. CEO Aaron Levie stated that advances in artificial intelligence are rapidly expanding the ability of businesses to automate complex professional tasks, as newer AI models become more capable of handling specialized knowledge work. Levie emphasized that the latest AI models are showing significant improvements in areas requiring advanced reasoning and industry expertise, particularly in sophisticated domains like legal, professional services and healthcare.

On Wednesday, in a post on X, Levie noted that "the latest AI models being dropped are getting insanely good handling complex knowledge worker tasks." He highlighted Grok 4.5 as an example of improving AI capabilities, specifically regarding cost and performance. Levie added that as models get better at coding, mathematics and reasoning, businesses will unlock new ways to utilize internal information and documents.

AI Adoption Reshapes Work

The shift toward advanced AI automation is visible across the enterprise sector. JPMorgan Chase & Co. CEO Jamie Dimon warned that AI could reshape the workforce by eliminating some jobs, urging preparation for potential disruption. Dimon said JPMorgan is already using AI widely, with about 150,000 employees using large language models weekly, while focusing on retraining workers affected by automation.

Former Google CEO Eric Schmidt said AI’s potential is being underestimated and could transform industries including finance, healthcare and scientific research. He argued that AI is a powerful business tool, helping automate tasks while creating opportunities for innovation. Meanwhile, Anthropic’s internal research on its Claude Code AI tool found that employees experienced higher productivity but raised concerns about reduced teamwork, declining technical skill development and uncertainty about future job roles. The study found workers could delegate up to 20% of their tasks to AI.