Leaked screenshots show Microsoft testing Copilot Canvas, an AI whiteboard

Leaked screenshots and developer endpoints shared on social media indicate Microsoft is internally testing a new AI-powered whiteboard called “Copilot Canvas,” also referenced as “Project Firenze,” that blends freeform drawing, note-taking and generative image tools inside a single web workspace. The material, posted “As posted on X by Windows enthusiast WalkingCat” and reported by Windows-focused outlets, includes references to both development and production Azure endpoints, a signal that the build may be active beyond static mockups.

The leaked interface shows a landing prompt reading “Create your first canvas to start drawing and taking notes,” a toggle labeled “Create with AI Streaming,” and an “Image Model Selector” suggesting multiple image-generation options inside the workspace. The interface also appears to autosave work, and includes a generic-looking logo that outlets said may not be final. WindowsReport journalist Milan Stanojevic published a writeup on March 1, 2026, describing the effort as part of “consolidating creative and collaborative tools under the Copilot brand” and situating it alongside other Copilot initiatives such as Copilot Tasks and Copilot features in Outlook.

Developer-focused controls exposed in the screenshots include a “Developer Mode” panel and items that outlets listed as Intent Detection, Meeting Summary, Solve Math, grounding options, and action delegation. Observers noted feature gates and toggles visible in the UI; in the words supplied by one source, “The presence of feature gates and developer toggles suggests the project remains in early-stage testing.” Multiple outlets emphasized that “Microsoft has not officially announced Copilot Canvas or provided a release timeline.”

The leak arrives against a backdrop of Microsoft’s existing Copilot tooling and Power Platform controls. Learn Microsoft documentation included in the leaked material underscores administrative limits: “Starting February 2, 2026, you can't add the Copilot control to new canvas apps. Existing apps with this control will remain functional for a limited time but will eventually no longer be supported.” The documentation also notes that “The Copilot control only supports Dataverse tables for the data source,” a constraint that matters for enterprise data architecture and governance.

Security and governance questions surfaced immediately in analysis pieces. Published fragments explicitly flagged risks including data exposure and access scope, streaming and hallucination hazards, autonomous actions and agent safety, IP copyright and provenance of generated media, and logging, telemetry, and regulatory compliance. Analysts also contrasted Copilot Canvas to today’s Microsoft Whiteboard and Copilot Pages while warning IT leaders to track file portability and enterprise flags.

Taken together, the leaked assets suggest Microsoft is exploring a persistent, multimodal workspace where generative AI moves beyond isolated chat assistants into collaborative canvases. For enterprises and admins, the immediate implications are practical: review data-routing and telemetry policies, verify compatibility with Dataverse constraints, and watch for governance controls tied to agent actions and streaming outputs. Reporters and security researchers can also point to the leak’s mention of development and production Azure endpoints as a high-priority verification item.

The development marks a clear push to fold creative and collaborative workflows into the Copilot umbrella. But until Microsoft confirms the product or provides a timeline, the project remains an internal test with significant unanswered questions about how the company will handle accuracy, provenance, and enterprise governance at scale.