Privacy and preparation
Dataset secure review screenshots: a screenshot is a new copy of restricted data
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
A secure data review should treat every screenshot, phone photo, copied text and verbatim note as a new copy of the restricted records. Reduce the routes with screen capture protection, watermarking and clipboard limits, then decide in advance which derived outputs may leave, in what form and who approves them. Log every release with a hash so both parties can later show exactly what left the session.
Checklist
Dataset secure review
Key takeaways
- A screenshot, a phone photo or a verbatim note is a derived export of the records on screen.
- Screen capture protection and clipboard limits cut casual copying but are not DRM, and nothing on screen stops a camera.
- Watermarks that encode the session make a photographed screen traceable.
- Decide which outputs may leave before the session and release them only through a logged, approved request.
- Viewing, keeping a derived copy and licensing are separate permissions; counsel reads the documents that grant each.
How should a secure dataset review treat screenshots?
Treat every screenshot as a new copy of the restricted records, created outside the controls you put on the originals. A review that forbids downloads but tolerates screenshots has only moved the export from a file to an image. The same goes for phone photos of the screen, copied text, verbatim notes and recorded calls.
The working answer has three parts. Reduce the routes a copy can take with workspace controls. Decide in advance which derived outputs may leave the session at all, in what form and on whose approval. Then keep evidence of both, so either party can later show what left the room and what did not.
This page is written for the data owner who hosts a review of records a buyer, an auditor or a processing partner needs to see before anything is delivered. It gives a reviewer workspace checklist and a derived export policy you can adapt, with one worked example and an exception. It is general information: the confidentiality and privacy duties that apply to your records depend on your agreements and the law that governs them, which counsel should read.
Which routes turn a viewing session into a copy?
Most leaks from a view-only review do not need any hacking. They follow ordinary features of the reviewer's device and the meeting tools in use. List them before the session, because each one needs a different answer.
| Route | How the copy is made | Control available | What remains |
|---|---|---|---|
| Operating system screenshot | Key combination or snipping tool on the reviewer's device | Client-enforced screen capture protection in a virtual desktop | Platforms and apps that do not honor the setting |
| Meeting screen share or recording | The reviewer shares the remote window into a call that records it | Capture protection that also blanks the window in screen sharing; meeting recording off by policy | A second participant's own device |
| Clipboard and file redirection | Copy and paste text or images out of the session, or save to a mapped drive | Clipboard direction and data type limits; drive and printer redirection disabled | Retyping by hand |
| Phone camera | A photo of the monitor | None technical; watermarking that identifies the session, plus signed terms | Any photo taken before someone notices |
| Background capture features | Operating system features that keep periodic snapshots of the screen | Device policy that disables them or excludes the review app | Unmanaged personal devices |
| Notes and AI note-takers | Verbatim record text typed or transcribed into a notes tool | Notes kept inside the workspace; note-taker bots barred from the session | Paraphrase that still identifies a person or account |
What can screen-capture controls block, and what do they leave open?
Screen capture protection reduces casual copying; it does not stop a determined reviewer. Microsoft's Azure Virtual Desktop documentation describes the feature as blocking remote content in screenshots and screen sharing, enforced by the client, and says plainly that it is not DRM-level protection. It can also block capture tools inside the session host. The same documentation recommends pairing it with disabled clipboard, drive and printer redirection and notes that some native macOS applications may not fully respect the setting, so Windows endpoints are preferred for the most sensitive content.
Nothing on the screen stops a camera. Microsoft's answer for that gap is deterrence: Azure Virtual Desktop watermarking overlays a QR code containing the connection ID or device ID, which an administrator can use to trace a photographed screen back to a session. Once watermarking is on for a session host, only clients that support it can connect.
Clipboard limits are finer-grained than on or off. Azure Virtual Desktop can restrict transfer from session host to client separately from the other direction, and by data type, from fully blocked through plain text only up to text, images, rich text and HTML. For a records review, block host-to-client transfer entirely unless a specific output has been approved.
Mobile platforms work the same way. On Android, FLAG_SECURE tells the system to treat a window's content as secure, keeping it out of screenshots and off non-secure displays. Android 14 added a callback that tells an app when the user takes a screenshot, but only for the hardware button combination; it is not invoked for windows with FLAG_SECURE set and does not detect screen recording. Detection tells you a copy was made; it does not undo it.
Background snapshot features are the route reviewers forget. Windows Recall stores periodic snapshots of the screen on the device. Microsoft's support pages describe filters for apps and websites and a sensitive information filter that is on by default, but they also say parts of filtered sites can still appear, for example through embedded content or a tab in the background. If a reviewer joins from a device where such a feature runs, decide whether the review app is excluded by policy or the device is not allowed.
Why viewing restricted records is already handling them
A screenshot policy is not only a security preference. Under the EU General Data Protection Regulation, processing means any operation on personal data, and the definition lists recording, storage, retrieval and consultation among its examples. Showing personal data to a reviewer, and the reviewer capturing it, are both operations on that data, so the review needs the same lawful basis, purpose limits and safeguards as any delivery. Other privacy laws frame this differently, and counsel should confirm what applies to your records.
Contracts usually reach the same place by another road. Confidentiality terms in customer agreements, vendor contracts and the review agreement itself commonly cover copies and notes as well as originals. Before a session, read which documents restrict the records on screen and whether those documents allow any third party to view them at all. Permission to view, permission to keep a derived copy and the right to license the records later are three separate questions, and a clean review answers only the first.
The reviewer workspace checklist
Set the workspace up before the first record appears, and keep the evidence each control produces. The checklist below assumes a hosted virtual desktop; adapt the settings column to the platform you use.
| Control | Setting for a restricted review | Evidence to keep | Owner |
|---|---|---|---|
| Named access | Individual accounts for each reviewer, no shared logins, access window that expires on a set date | Account list with start and end times | Data owner's administrator |
| Screen capture protection | On for the session host, including capture tools inside the session | Policy export with the setting and the date applied | Administrator |
| Watermarking | On, showing the connection ID | Session connection IDs mapped to reviewer names | Administrator |
| Clipboard and redirection | Host-to-client clipboard blocked; drive, printer and USB redirection off | Policy export | Administrator |
| Endpoint rules | Managed or approved devices only; background snapshot features disabled or the client excluded | Device attestation or signed reviewer statement | Reviewer's organization |
| Records on screen | Smallest sample that answers the review question, pre-redacted where possible | Sample definition and the record IDs shown | Data owner |
| Notes | Kept in a notes file inside the workspace, never in an external tool | The notes file, retained with the output register | Reviewer |
| Session conduct | No meeting recording, no note-taker bots, camera use outside the room agreed in writing | Signed review terms | Both parties |
The derived export policy: what may leave the session, and on whose approval
Decide what the reviewer may take away before they arrive, and route every exception through a named approver. Secure research services use the same principle. The UK Data Service SecureLab, for example, says users can view and analyze data remotely but may not copy analytical outputs from the screen, and that outputs are released only after staff carry out statistical disclosure control checks. Researchers are also asked to check for secondary disclosure, where two outputs that are safe alone reveal something together. A commercial records review rarely needs statistics, but it needs the same gate.
- Log every output request with the requester, the date, the decision, the approver and a hash of what was released.
- An output that was refused stays refused unless the scope or the agreement changes; record which.
| Output the reviewer wants | Default | Condition to release | Approver |
|---|---|---|---|
| Screenshot of a record | Not released | Replace with record ID plus a redacted excerpt produced by the data owner | Data owner's privacy lead |
| Screenshot of the interface with no record content | Released on request | Checked for names, account numbers and other identifiers in menus, headers and sidebars | Data owner |
| Counts and rates from the sample | Released | No cell small enough to point at one person or account; check combinations across outputs | Data owner |
| Reviewer notes | Released after review | Verbatim record text removed; findings cite record IDs only | Data owner's privacy lead |
| Example records for a scoping document | Not released | Only after the agreement is signed, as part of an approved delivery | Counsel and the release approver |
Worked example: a label-quality review that wanted screenshots (illustrative)
This example is invented to show how the checklist and policy work together. It does not describe a real company or review, and all counts are illustrative.
A data owner hosts a two-day review of 300 support tickets for a prospective buyer's annotation lead, who is checking whether resolution codes match the conversation text. The workspace follows the checklist: capture protection, watermarking, no host-to-client clipboard, notes inside the workspace.
- Day one: the reviewer finds 14 tickets where the code contradicts the thread and asks to keep screenshots for the internal report. The policy says no. The data owner produces a short table of the 14 record IDs, the stated code, the code the thread suggests and a redacted one-line excerpt for each. The reviewer's report cites that table.
- Exception: the reviewer also asks for a screenshot of the ticket form layout to explain where the code field sits. The data owner clears a capture of the empty form after checking the sidebar for agent names, logs the release and hashes the image.
- Incident: on day two a colleague of the reviewer joins and photographs the screen with a phone to share in a chat. The watermark identifies the connection, the session is ended and the photo's deletion is confirmed in writing with the time and the chat where it was posted. The incident and its handling go into the output register.
- Close: the notes file is reviewed, three verbatim customer sentences are removed, and the cleaned notes are released. Access expires on schedule.
How both parties confirm the access limits held and what actually left
Both sides should be able to show the same picture of the session from their own records. The host can show policy exports, the account list, connection IDs and the output register with hashes. The reviewer's organization can show which devices joined, a statement that no meeting recording or background snapshot ran, and a confirmation that any unapproved copy was deleted, with what was deleted and when.
Compare the two at close. Every released output should appear in both registers with the same hash. Any copy that appears on one side only is an incident to resolve before anyone discusses a delivery. Keep the comparison with the review file; if the records are later licensed, it shows that nothing left the review except what was approved.
When the fix is fewer records on screen, not tighter controls
If the review question needs hundreds of identifiable records on screen, the controls are carrying too much weight. Ask whether the question can be answered with fewer or less identifying records first. Label quality, field completeness and template versus free text can usually be judged on a pre-redacted sample. Distribution questions can often be answered with counts the data owner runs.
Stop and escalate when the records include health, financial account or children's data, when a customer contract restricts any third-party viewing, or when the reviewer's organization cannot confirm device controls. Those are cases for counsel and the privacy lead to decide the format of the review, not for the administrator to solve with another setting.
Setting review terms before records are shared with SourceX
The order of steps matters here. SourceX receives no data until the company has approved scope and terms and signed the agreement, so records never travel to SourceX while the review question is still open. That leaves the data owner in control of what is shown during scoping and how derived outputs are handled.
Seeing whether a company qualifies starts with a short assessment and needs no data upload, so nothing has to appear on anyone's screen at that stage.
Frequently asked questions
Can screen capture protection stop all screenshots?
No. In Azure Virtual Desktop it blocks screenshots and screen sharing of remote content through the client, but Microsoft says it is not DRM-level protection, some macOS apps may not fully respect it, and it cannot stop a phone camera. Combine it with watermarking, clipboard and redirection limits, and signed review terms.
Is a screenshot of anonymized records still a problem?
It can be. Pre-redacted records shown on screen still carry context, and several screenshots together can identify a person or account even when each one alone does not. Treat screenshots as outputs under the derived export policy and release record IDs and redacted excerpts instead.
Should reviewers be allowed to take notes during a secure data review?
Yes, but inside the workspace. Notes typed into external tools or captured by AI note-takers leave with whatever record text they contain. Keep notes in the session, review them at close, remove verbatim record content and release the cleaned file through the output register.
How do you trace a photo of a screen back to a session?
Use a visible watermark that encodes the session. Azure Virtual Desktop watermarking overlays a QR code containing the connection ID or device ID, which an administrator can map to the reviewer who held that connection. Keep the mapping with the review file.
Does a view-only review give the reviewer any right to use the data?
No. Viewing under review terms answers only whether the reviewer may see the records for that purpose. Keeping a derived copy, using records for training and licensing them are separate permissions that come from signed agreements, which counsel should read for the specific records.
Sources
- Microsoft's Azure Virtual Desktop documentation says screen capture protection blocks remote content in screenshots and screen sharing, is enforced by the client, is not DRM-level protection, and should be paired with disabled clipboard, drive and printer redirection. Source
- Azure Virtual Desktop watermarking overlays a QR code containing the connection ID or device ID so administrators can trace a session, and only supporting clients can connect once it is enabled. Source
- Azure Virtual Desktop can restrict clipboard transfer from session host to client by data type, from blocked to plain text up to text, images, rich text and HTML. Source
- Microsoft's Recall support page describes app, website and sensitive information filters, the sensitive information filter on by default, and cases where parts of filtered websites can still appear in snapshots. Source
- Android's FLAG_SECURE treats a window's content as secure, preventing it from appearing in screenshots or being viewed on non-secure displays. Source
- Android 14 screenshot detection reports only screenshots taken with the hardware button combination and is not invoked for windows with FLAG_SECURE. Source
- The UK Data Service SecureLab FAQ says users may not copy analytical outputs from the screen and that outputs are released only after statistical disclosure control checks by service staff. Source
- GDPR Article 4(2) defines processing as any operation on personal data, including recording, storage, retrieval and consultation. Source
See if your company qualifies
A short company assessment. No data uploads are needed.