Decentralised News · 2027 planning series · Reviewed 22 November 2026
What Matters
A private login can reduce what an application learns when you enter a virtual space. The rest of the session needs its own privacy design. An age-threshold proof, for example, says little about which application receives your movements, whether an analytics service stores them, or whether another visit uses the same account identifier.
For a VR product planned for 2027, follow both paths: the credential presented at admission and the data produced while the experience runs. The useful decision is which input the feature actually needs, where that input goes and when each retained copy disappears. This article connects a named WebXR implementation to that decision and supplies a local dataflow record. It does not claim that a particular headset integrates nine zero-knowledge features or that DN tested an anonymous VR deployment.
The Signal: One Platform Can Offer Different Input Boundaries
WebKit’s Safari 18 release notes document immersive WebXR on visionOS 2, including transient-pointer interaction and separately permitted hand tracking. Those are concrete implementation choices. A product can therefore ask a narrower question than whether a platform is generally private: does this interaction need continuous hand data, or will an input event do?
The release is dated September 2024. It establishes a shipped example, not a test of every later browser build. Keep the actual browser, operating-system version and requested features with any deployment review.
Start With the Claim the Login Actually Reveals
Consider a hypothetical age-restricted virtual exhibition. Its admission service needs a decision about whether the visitor meets an age threshold. A design might disclose a full date of birth, a signed age-band claim, or a proof of a more limited statement. These are alternative designs, not descriptions of an integration that DN has deployed.
The W3C Verifiable Credentials Data Model 2.0 describes an issuer, holder and verifier ecosystem and discusses selective disclosure. It also identifies correlation through identifiers, signatures and metadata. A credential format alone does not establish that every presentation is unlinkable, and selective disclosure should not automatically be labelled zero knowledge without identifying the actual securing mechanism.
DN recommends writing the admission claim as a sentence: “This presentation establishes the stated age condition for this access request.” Then list the information actually delivered alongside it. A session token, account name or transaction record can belong to a separate part of the design. If the service receives one of those identifiers, explain its purpose and lifetime instead of allowing the word “private” to stand in for a data contract.
The question is not whether the credential is useful. It is which problem it solves. A smaller admission disclosure remains valuable even when a later feature needs motion data. The design should account for both decisions.
Follow the Session Beyond the Admission Service
The 9 November 2026 WebXR Device API draft treats sensor-derived poses as sensitive information. Immersive rendering needs pose information, and the specification describes fingerprinting and profiling concerns. Its pose-reporting algorithm excludes hidden sessions. This is a Candidate Recommendation Draft, not a final W3C Recommendation or proof that every browser implements each protection identically.
For a product review, draw a boundary around each recipient. The headset runtime, page code, game server, voice service and analytics endpoint should not be collapsed into one box called “the platform.” Some components may be absent from a particular product. The record should describe the actual design, including the components whose behavior is still unknown.
A feature can use information locally without requiring the same raw stream to be retained remotely. Whether that separation is achievable must be established for the feature and implementation. Treat “needed to render this frame,” “needed to synchronize another participant” and “wanted for future analysis” as three different purposes. Each deserves its own justification and evidence.
Read the Named Input Mechanism Closely
WebKit’s natural-input explanation describes a transient input created when a user pinches. Its target ray initially follows the gaze direction at the start of the gesture; subsequent movement follows the hand. The input disappears when the interaction ends. Separately requested and granted hand tracking supplies persistent hand-joint inputs for the experience.
That distinction does not mean the transient path reveals nothing. It communicates an interaction with a location and timing. It also does not mean the page receives an unrestricted continuous eye-tracking feed. Review the specific interface rather than substituting either extreme.
Apple’s native visionOS guidance offers a related, separate example: system input can deliver interactions without directly passing continuous eye and hand positions to an app. Direct hand-position access requires authorization and an immersive space. Native app interfaces and browser interfaces must remain distinct in the review.
For the hypothetical exhibition, selecting a door and manipulating a hand-attached object may have different input requirements. Ask the team to demonstrate the intended interaction with the least revealing supported input that meets the task. Record the result as a product-specific finding. Do not turn a platform document into a universal promise about every application built on it.
Motion Can Link Sessions Without Revealing a Name
A USENIX Security 2023 study by Nair and colleagues examined 55,541 VR users. With five minutes of training data per person, its classifier identified users within that study pool with 94.33% accuracy from 100 seconds of head and hand motion, and 73.20% from ten seconds. The paper used repeated-session Beat Saber data and explicitly limited what it demonstrated across applications.
This is evidence of linkability under the study’s conditions. It is not a probability that any website can discover an arbitrary visitor’s real name, and it is not a result DN reproduced. The operational implication is narrower: omitting a name from a session record does not establish that repeated motion records cannot be associated with one another.
DN would therefore review a proposed motion archive before treating it as anonymous. The relevant questions include who can compare records, which reference data they possess and whether an account or session identifier already supplies a direct link. Those questions concern the particular archive; the published accuracy figure should not be pasted into a product’s privacy score.
Give Each Data Path Its Own Record
The following structure is DN’s proposed review method. Its rows are questions for a deployment team, not findings about Safari, Apple, a credential issuer or a game operator.
| Path to examine | Decision to record | Evidence needed |
|---|---|---|
| Credential to admission service | Exact disclosed claim and accompanying identifiers | Protocol, presentation fields and verifier configuration |
| Runtime to application | Input events or continuous streams actually exposed | Version, requested features, permissions and observed API behavior |
| Application to another service | Purpose and minimum fields for each recipient | Message schema and a permitted inspection of the relevant data path |
| Service to retained record | Retention rule, linkage and access | Storage configuration, responsible owner and deletion evidence |
Separate permission from collection. An allowed capability may go unused. A planned flow may never be enabled. Conversely, a record that claims collection while its permission state is denied needs investigation. The worksheet preserves those states instead of declaring that a permission setting proves what happened.
Use one row for each data, recipient and purpose combination. If the same pose stream reaches both a local renderer and an analytics service, enter two rows. Their source may be shared while their retention and access differ. Do not mark a whole application “session only” because one component discards its buffer when the view closes.
Retention Needs an Owner and an Observable End
DN recommends distinguishing transient use, retention until session end, a fixed retention period, indefinite retention and an unknown rule. These are declarations about a specific copy, not grades of safety. A short-lived record can still contain sensitive information or reach an inappropriate recipient.
For a fixed period, define when the clock begins. “Seven days” could mean seven days after collection, after the session ends or after a later support ticket closes. Keep that trigger in the underlying record. The worksheet accepts a period label but does not enforce a deletion job or establish which trigger the service uses.
Derived records deserve a separate row when they survive the raw data. An aggregate, a behavioral embedding or a troubleshooting extract may have another owner and lifetime. Deleting the original stream does not, by logic alone, establish deletion of every result previously made from it. Ask which transformation produced the retained object and what evidence addresses that object.
Apply the same discipline to an AI assistant inside the experience. If the design sends voice transcripts, interaction events or movement summaries to a model service, describe those fields and that recipient. This is a proposed review question, not a claim that a named VR platform sends any particular telemetry to a model provider.
Assign a person or team to resolve each unknown. “Analytics configuration not reviewed” is more useful than “privacy approved” when nobody has located the destination or retention setting. The next action can then be precise: inspect one integration, remove one unnecessary field or confirm one deletion path.
A Worked Design Review for the Virtual Exhibition
Suppose a fictional team proposes three flows. Its admission service receives an age-condition result. Its local scene uses head pose while rendering. A prospective analytics integration may receive hand-joint records, but the team has not established its permission behavior, retention rule or cross-session identifiers. These are synthetic design declarations, not observations of a real application.
The team records the first two paths as planned, with evidence aliases pointing to design documents. A planned transient buffer is still a proposed configuration, not a measured absence of storage. The analytics row remains unknown. The admission proof cannot resolve it because the proof concerns a different data path.
Now ask what the analytics feature is meant to answer. If the requirement is simply to count successful door selections, the team can evaluate an event-level design instead of assuming it needs a movement archive. This is a design alternative to test. It is not a claim that event data is automatically anonymous or that every interaction can be implemented with the same interface.
A useful comparison keeps the feature objective constant. Can the revised input complete the task? What changes for accessibility, multiplayer coordination or error diagnosis? Which additional data would an exception require, and for how long? A privacy review should expose those tradeoffs clearly enough for a product owner to choose.
If the team later observes a hand stream reaching a service, it changes that row from unknown to observed and retains the observation reference. If the recipient’s deletion behavior remains unknown, that cell stays unknown. Better evidence in one column should not silently fill the others.
Finally, suppose the team disables collection for a future release. It records that future state separately from any already retained copy. A new configuration can prevent another upload without proving that yesterday’s data disappeared. Keep the old-copy question open until the relevant evidence exists.
Before rollout, define an authorized verification exercise using suitable test data. Name the feature, version, expected recipients and permitted observation method. Include a permission-denied case, a session ending normally and the retention trigger that the design promises. A reviewer should be able to distinguish the expected result from the observation, the observation time and the unresolved limitation.
A successful exercise has a bounded meaning. Seeing no upload during one interaction supports that observed case; it does not establish that every scene, error path or later configuration behaves the same way. A deletion test should identify the particular stored object and recipient examined. Preserve those limits when moving a worksheet row from planned to observed.
Use the DN VR Session Dataflow and Retention Record
Enter a nonpersonal review alias, a platform/version alias and the limited login claim. Then enter one row per data path using the displayed column order. The initial rows represent the fictional exhibition. Replace them with your own documented design or observations; do not paste raw motion samples, credentials, names or account identifiers.
For every row, distinguish permission, collection, retention and cross-session linkage. “Unknown” is an accepted state. A blank or malformed row is an input error rather than evidence of no collection. An evidence alias points to a record you keep elsewhere; the worksheet does not open or authenticate it.
The output counts entered flows and unresolved categories, flags a stated permission conflict and lists the next questions. Those counts organize work. They are not a probability of identification, a comparison of vendors or a legal compliance result. Even a fully populated record remains a set of declarations requiring the appropriate verification.
DN VR Session Dataflow and Retention Record
Change the illustrative assumptions below. This worksheet runs locally in your browser. It records your declarations; it does not verify evidence, connect to accounts, submit instructions or grant approval.
The initial numbers are synthetic. Unknown evidence remains unresolved. Downloads contain the values you enter; use aliases and avoid private or identifying information.
DN Alpha Thesis
The useful design advantage is a smaller, explainable data path for each feature. A private admission proof and a restrained interaction interface can serve different parts of that design. Their value becomes assessable when the team can say what each recipient receives, what it keeps and which question is still unanswered.
For 2027 planning, make that record part of a feature decision. If an integration changes the receiver, input granularity, identifier or retention period, revisit the affected rows. An unchanged login proof should not conceal a changed session data contract.
Methodology, Limits, Corrections and Disclosure
DN reviewed the linked W3C, WebKit, Apple and USENIX materials on 22 November 2026. The dated platform releases establish documented interfaces and history. The current WebXR document is a draft. No headset, browser session, credential presentation, network transmission or deletion process was independently tested for this article.
The exhibition scenario and dataflow worksheet are original DN analysis. They compare declared purposes and states, not biometric measurements. The tool makes no network requests and does not activate a sensor, connect a wallet or evaluate a cryptographic proof. Its exported record contains the values entered; use aliases and retain sensitive evidence in the appropriate system.
Related DN research examines personhood data representations and locations, credential status and freshness, and agent data across models, APIs and logs. This article adds the separate question of continuous immersive-session inputs. It supplies no nine-product ranking, anonymity certification or legal determination. Corrections should identify the affected source, implementation version and data path so the corresponding claim and worksheet guidance can be reviewed together.
Sources and evidence register
Primary sources reviewed on 22 November 2026. The article distinguishes documented facts, DN analysis and synthetic examples. Recheck changeable rules and configurations before acting.
- W3C WebXR Device API Candidate Recommendation Draft — 9 November 2026 Candidate Recommendation Draft; sensitive poses and session reporting boundaries, not a final Recommendation or implementation test.
- WebKit: Introducing Natural Input for WebXR in Apple Vision Pro — 19 March 2024 implementation description; transient gesture input and separately granted persistent hand tracking.
- WebKit Features in Safari 18.0 — 16 September 2024 shipped Safari 18/visionOS 2 capabilities; a dated release, not a test of later builds.
- Apple visionOS: Adopting best practices for privacy — Native visionOS input and authorization guidance; not interchangeable with WebXR interfaces.
- Nair et al., USENIX Security 2023: identification from head and hand motion — 2023 repeated-session study, 55,541 users; closed-pool identification results, not arbitrary real-name identification or DN testing.
- W3C Verifiable Credentials Data Model 2.0 — 15 May 2025 Recommendation; credential claims, selective disclosure and correlation considerations.
Frequently asked questions
Does a zero-knowledge login hide head and hand movement?
A proof addresses its specified claim and disclosure protocol. Session inputs use a separate data path. Identify which motion information reaches each recipient instead of extending the login claim to the whole experience.
Does the documented transient-pointer mode expose continuous gaze?
WebKit describes a gaze-directed starting ray for a pinch, then hand-driven movement during that interaction. That is different from an unrestricted continuous eye-tracking feed. Separately granted hand tracking has another exposure scope.
What does the study’s 94.33% identification result mean?
It is identification within the 55,541-person study pool using 100 seconds of motion after five minutes of training data per person. It does not establish arbitrary real-name identification or universal tracking across applications.
Is an interaction event automatically anonymous?
No such conclusion follows from the event format alone. Review its identifiers, recipient, retained content and relationship to other records. Less data can be useful without establishing anonymity.
Does a fully populated worksheet prove the application is private?
No. It records declarations and organizes unresolved questions. It does not inspect sensors, network traffic, evidence references, storage or deletion, and it supplies no compliance certificate.
Which version should a 2027 deployment review use?
Record the actual browser, operating system, requested features and service configuration used by the deployment. Keep dated release evidence and draft specifications separate from observations of that specific build.