How to Approve AI Plugins for Revit, AutoCAD and Bluebeam

AI plugins for Revit, AutoCAD and Bluebeam are reaching architecture and engineering firms faster than most principals can review them. A designer finds an AI add-in that promises to tag a Revit model in minutes. A project manager wants an AI assistant connected to Bluebeam so it can summarize markups. Both tools run inside software that holds your clients' drawings, and both can send project data to a cloud service the firm has never reviewed. This checklist covers what to check before approving an AI plugin, how each program loads plugins, and how to keep unapproved ones off design workstations.

Why do AI plugins for Revit, AutoCAD and Bluebeam need their own approval step?

They need their own step because a plugin runs inside the design application with access to whatever is open, and many AI plugins send that content to an outside service. A firm-wide AI policy that approves a chatbot says nothing about an add-in that reads an entire Revit model. For a principal, the exposure is client drawings reaching a vendor the firm never agreed to share them with, which can conflict with confidentiality terms in owner and client agreements.

Bluebeam's documentation for its Revu MCP connection is a useful example of how specific the data flow can be. The connection lets an outside AI assistant run workflows in Revu 21.9 or later on a Bluebeam Max plan. Bluebeam states that the MCP server runs locally on your machine, but that "regardless of the privacy settings you've selected, prompt and text data are transmitted to the AI model's servers." Bluebeam adds that training depends on the AI model you connect, and that with a cloud provider you have to configure its privacy settings yourself to prevent training. The approval decision covers two vendors: the design software publisher and the AI provider behind it.

What should you check before approving an AI plugin?

Check the publisher, the data flow and the off switch before anyone installs the plugin. A short review covers most of the risk:

• Publisher. Who wrote it, how long they have supported it, and whether it is distributed through the software vendor's official app store or a direct download.
• Code signing. Autodesk's security guidance for AutoCAD says to confirm that third-party executable files are digitally signed by a reputable issuer.
• Data flow. What leaves the workstation: geometry, sheet text, markups, file names, project metadata. Where it is processed, and by which AI provider.
• Training and retention. Whether the plugin publisher or its AI provider trains on your content, how long they keep it, and whether that differs between free and paid tiers.
• Account type. Whether the plugin uses a firm-managed account or each user's personal AI account. Personal accounts put the training setting in the user's hands.
• Off switch. Whether an administrator can disable the feature centrally, and whether it can run with a local model instead.
• Client terms. Whether any client agreement restricts sharing project information with third parties.

Write the answers down with the date and the version reviewed. Approve the plugin for named user groups and project types instead of the whole firm.

How do Revit, AutoCAD and Bluebeam load plugins?

Each program loads plugins differently, which tells you where to look and what to lock down. If you are also planning hardware for AI features, see what Revit workstation to buy for rendering and AI.

Revit registers add-ins through .addin manifest files. Autodesk's add-in registration guide lists two locations: an all-users folder under C:\ProgramData\Autodesk\Revit\Addins\ and a per-user folder under the user's AppData\Roaming\Autodesk\Revit\Addins\, each with a subfolder for the Revit version. A user who can write to the per-user folder can add a plugin without IT. Inventory both folders on every workstation.

AutoCAD controls loading with the SECURELOAD and TRUSTEDPATHS system variables. Autodesk recommends setting SECURELOAD to 1 or 2 to prevent unauthorized code from running, keeping executable code in trusted, read-only folders separate from data, and not running AutoCAD with administrator privileges.

Bluebeam's AI features are managed through account and application settings. Bluebeam's AI transparency page says administrators can disable Intelligent Studio Project Search (with Studio), Smart Overlay, Smart Review and the MCP connection in Org Admin Pro, while OCR, Visual Search, Auto Align and Stitching cannot be disabled. That was the list as of October 2026, so check the page before each Revu rollout. Decide which of the optional features your firm allows and set them centrally.

What about AI features built into software you already license?

Treat them as plugins that arrived with an update. Autodesk publishes AI transparency cards that describe each AI feature's data sources, model provider and safeguards, and note whether customers can turn a feature off at supported levels or whether it is always on. Read the card for each feature before rolling out a new version.

Autodesk introduced Autodesk Assistant in Revit as a tech preview tagged to Revit 2027. Autodesk states that none of the data the assistant works with, nor any customer privacy data, is used to train LLM models, and that prompt libraries and chat history are stored in the cloud and linked to the user's Autodesk account. Those are the kinds of facts your approval record should capture for every built-in feature.

How do you keep unapproved AI plugins off design workstations?

Use the same software controls you apply to any other application. The CIS Critical Security Controls call for an inventory of authorized software (Safeguard 2.1), removal or documented exception for unauthorized software (Safeguard 2.3) and technical controls such as allowlisting so only authorized software can run (Safeguard 2.5). Our guide to CIS Controls 1 and 2 asset inventory covers how to build that inventory.

In practice, for a design office:

• Remove local administrator rights from designers' everyday accounts.
• Deploy approved plugins centrally to the all-users locations, and alert on new files in the per-user add-in folders.
• Set AutoCAD's SECURELOAD and trusted paths through your deployment, not by hand on each machine.
• Turn on new-app alerts so sign-ups for AI services with firm email reach IT.
• Block unapproved AI sites at the web filter so a plugin that calls one fails closed.

Can AI plugins be used on confidential or defense projects?

Only when the plugin and its AI provider have been approved for that specific project. Security-sensitive facilities, client agreements with confidentiality clauses and defense work with controlled unclassified information all carry stricter limits than a typical commercial project. On defense work, any cloud service that handles covered defense information has to meet the bar in DFARS 252.204-7012, so confirm in writing that a plugin and its AI provider meet it before either touches a defense project. Until then, keep AI plugins off those projects.

What should your firm do this quarter?

• Inventory the add-in folders and installed plugins on every design workstation.
• List the AI features in your current Revit, AutoCAD and Bluebeam versions and decide which to allow.
• Review each AI plugin in use against the checklist above, and remove the ones that fail.
• Set SECURELOAD, trusted paths and Bluebeam Org Admin Pro settings centrally.
• Remove local administrator rights where designers still have them.
• Publish a short approved-plugin list so staff know what they can use.

How does s90 help?

s90 inventories the software on your design workstations, sends new-app alerts when someone installs a plugin or signs up for an AI service, and sets AI guardrails so approved tools run on firm-managed accounts. We keep plugin reviews in the 18-month technology plan we maintain with each client. If your designers are asking for AI plugins and you want a review process that does not slow them down, we can set one up with your team.