Proving a Human Reviewed It with AI Attribution and Watermarks in Board Reporting
The board does not care whether AI helped build the deck. It cares whether the numbers are right and whether a competent human stood behind them. That is a harder assurance to give now that Copilot is editing workbooks alongside your team, because the old proxy for accountability, "a person typed this," no longer holds. When a model and three analysts have all touched a file, who reviewed what becomes a real question, and for a board pack it is a question you need to be able to answer.
Microsoft has started shipping the provenance tooling to answer it. The [Show Changes experience for Copilot in Excel](https://techcommunity.microsoft.com/blog/excelblog/whats-new-in-excel-may-2026/4502700), introduced in May, now flags AI-assisted edits explicitly: when a collaborator makes a change with Copilot, the Show Changes card carries a Copilot attribution indicator, a small visual marker and icon showing exactly where the model contributed rather than a person. That sounds cosmetic until you are the one signing off a variance schedule and need to know which cells came from an AI edit versus a preparer's judgement. Attribution turns "the workbook changed" into "here is what changed, and here is who or what changed it."
Alongside this, Microsoft has been rolling out [watermarking controls for AI-generated content](https://learn.microsoft.com/en-us/microsoft-365/copilot/release-notes) across Microsoft 365, letting organisations mark content that was generated or altered with AI so it cannot be quietly misattributed. The direction of travel is clear: provenance is becoming a first-class feature of the tools, not something you bolt on afterward.
The practical discipline for board reporting is to use these markers as the backbone of a review trail rather than a curiosity. Let attribution show where AI touched the model, then require a named human to review and clear those specific changes before anything reaches the deck. The goal is not to hide that AI helped; that ship has sailed and there is nothing wrong with the help. The goal is to be able to demonstrate, cell by cell if asked, that a person applied judgement to every number the machine produced.
At Cell Fusion Solutions Inc we wire attribution and sign-off into the reporting workflow so the provenance of each figure is defensible by the time it hits the board table. Show the AI's fingerprints, then show the human's.
