Box has unveiled Box Architect, a new tool that builds, tests and deploys agentic workflows from a plain-language description, using the content and permissions a company already has in Box. Announced November 5, 2026 at BoxWorks 2026 in San Francisco, Architect arrives alongside new threat protection for agent-era attacks and agent security controls that are available now. Why it matters: Box is betting the next enterprise AI contest is not chat — it is governed, multi-step work on real business documents.
November 6, 2026
What happened
Box, Inc. (NYSE: BOX) said Box Architect will let teams create workflows and apps, configure agents, connect Box tools and manage workflows at scale, all built on trusted content and existing permissions. It draws on Box Extract, Box Automate, Box Apps, Box Forms and other Box capabilities to turn a described process into a unified, repeatable workflow with people kept in at key decision points.
Customers can choose a model for each step from providers including Anthropic, OpenAI, Google, NVIDIA and IBM. Content stays in Box, governed by existing permissions, classifications and retention policies.
How Architect is meant to work
Box describes three stages. Teams design a workflow by describing a process or goal in plain language, and Architect uses content and procedures already in Box to define stages, decisions and handoffs. They test it before it goes live using real content and edge cases, previewing agent outputs and routing and adding human review at key decisions. They then deploy the complete workflow with the apps, agents, automations, forms and templates it needs, while Architect monitors results and recommends improvements.
Box’s examples include contract review that routes high-risk clauses to attorneys, bank client onboarding through Forms, Extract, Doc Gen and Sign, and government and insurance workflows for permitting and claims.
Security, and when you can get it
Box is also building threat protection to detect bulk exfiltration, counter session hijacking and account takeover with session binding, identify suspicious sharing and mass deletion, and apply automatic severity-based responses. Its agent security controls — classification-based access limits, MCP Guardrails and threshold alerts — are now generally available.
On timing: Box Architect will begin rolling out in beta in Q4 to Enterprise Advanced customers. The new threat protection capabilities are planned for Enterprise Advanced and Shield customers early next year, and the agent security controls are now generally available to Shield and Enterprise Advanced customers.
Why it matters
For subscribers, the signal is where enterprise AI budgets are going: away from isolated chatbots and toward orchestration layers that sit on top of a company’s documents, with model choice per step and governance built in. The honest caveat is that Architect is a beta roadmap item, not a generally available product today — the part customers can actually use now is the security controls. Everything AI, tested means separating those two timelines.
Frequently asked questions
When is Box Architect available?
Box says Architect will begin rolling out in beta in Q4 to Enterprise Advanced customers. No specific date was given.
Which AI models can Architect use?
Box says customers can choose models from providers including Anthropic, OpenAI, Google, NVIDIA and IBM, selecting a model for each workflow step.
What Box AI security is available today?
Box’s agent security controls are now generally available to Shield and Enterprise Advanced customers. The new threat protection capabilities are planned for early next year.
Sources: Box, Inc. via Business Wire, MarTech Series.