Skip to content
problem genome

One of 592 unsolved problems in the Problem Genome, a collection written up for student design and engineering teams.

digital · play · social

a bubble that keeps every stranger out

Social VR platforms give players mute, block, report and personal-space bubbles, but in a 100-player survey harassment stayed common and the VR-specific tools were little used; for disabled players, reactive tools act too late, default boundaries also keep out welcome strangers, and visible tools carry a social cost

Source: Weerasinghe et al., “Beyond Mute and Block: Adoption and Effectiveness of…” Open ↗source: primaryLast updated: 2026-11-17

Problem statement

Social virtual reality (social VR) apps such as VRChat, Rec Room and Meta's Horizon Worlds are places where people play together as full-body avatars: games, hangouts, dances, quizzes. Because the avatars are embodied, harassment can be physical: an avatar blocks your path, crowds you, or "touches" you. Platforms have added safety tools. Some come from older online games (mute, block, report, vote-kick). Some are new to VR: personal-space bubbles that keep other avatars at a distance, interaction shields, safety gestures. A University of Glasgow team reviewed the tools on 12 popular social VR platforms and surveyed 100 players from 22 countries (mean age 27). Harassment was common: 80% had met verbal harassment, as victim or witness. The tools players actually used were the old reactive ones (mute and block, 61%), which act only after harm is done. The VR-specific tools were rated effective by those who had tried them, but far fewer had: personal-space (proxemics) settings 25%, interaction shields 18%, safety gestures 16%, which the authors suggest may be because many platforms lack them. Players also used the exclusion tools to keep strangers out altogether, which the authors say fragments shared play spaces. For disabled players who show their disability on their avatar, the trade-off is sharper. Meta's Personal Boundary keeps strangers (people you don't follow back) at a distance by default; it can be set to everyone or off, but not to "strangers who ask first". The authors of a 2026 co-design study with 11 disabled adults argue that it "blocks all forms of social interactions, even though some PWD welcome engagement with their avatars with disabilities upon consent" (citing an earlier diary study). Their participants also held back from visible, confrontational tools because of what using them signals to others. The unsolved problem is a protection that stops harm before it lands without shutting out the welcome contact that is the point of playing in social VR.

Why this matters

In the Glasgow survey, 44% of players had been victims of hate speech, 27% had had their avatar's movement deliberately blocked, and 11% reported virtual sexual assault. Many described harassment as simply part of social VR. One said it made them feel unsafe "that's why I rarely use those platforms". Reporting was among the lowest-rated tools and the most opaque ("I report a lot of people but nothing ever happens"). In a separate survey of 67 disabled social VR users, the ASSETS '26 team identified 19 types of harm in 5 categories, and noted that visible disability signs on avatars can attract bullying, ableist language and insults. Many people with disabilities show their disability on their avatars to build social connections and foster meaningful conversations, so a tool that keeps everyone away protects them by taking away what they came for.

What’s been tried and why it hasn’t worked

Reactive tools (mute, block, report, vote-kick). These are on almost every platform, and players know them from other games. They work after the event: "usually all the damage has already been done" (survey participant), and a blocked harasser can return on "another account and another and another". Reports were described as slow and opaque. Since then some platforms have added automated proactive moderation: Rec Room runs ToxMod voice moderation, and its vendor reports large drops in repeat violations after warnings and mutes, without a stated method or an independent evaluation; VRChat announced in 2025–2026 more Trust & Safety spending, faster report handling and a test of a toxicity system in public instances, with no results published. (Automated moderation itself belongs to the collection's AI-evaluation area and is not this brief's subject.) In the Glasgow survey, only 21% of respondents were comfortable with AI moderation, against 67% for human moderation, and an earlier study the Glasgow paper cites found that human moderators addressed only 24% of incidents.

VR-specific tools, most of them proactive. Personal-space bubbles, interaction shields, safety gestures and trust-based filters were rated effective by 60–92% of the players who had used them. But they are unevenly available: safety gestures and trust ranking exist on only one major platform each. They are also less visible, and one user said safety gestures were "too slow, it's quicker to pull up a users profile and blocking them". A 2022 CHI workshop review of four platforms had already catalogued their boundary settings and quick-reaction tools and called for boundary-aware design; a 2023 study of four platforms, as the Glasgow paper reports it, found that most safety features are reactive and that relying on proximity alone is insufficient. Some features let players admit friends and keep strangers out ("disallow users I don't know away from me while letting my friends get close"). That works for friends, but it is the same exclusion the Glasgow authors link to fragmented social spaces.

Personal Boundary (Meta, 2022). At launch it was always on for everyone ("an almost 4-foot distance"). Since March 2022 it is on by default for "people you don't follow back", with "On for Everyone" and "Off" as the other settings; when two people meet, the more restrictive setting applies. In December 2023 Meta made intruding avatars fade faster and loosened the barrier for high fives. So the boundary is graded by friendship, but for a stranger it is still on or off. The co-design authors argue that such a boundary blocks welcome engagement too (a stranger who asks, with consent, about a wheelchair or cane avatar), drawing on earlier work. Their own participants avoided visibly confrontational tools, and some answered mockery with humour to present themselves as relaxed. The authors write that participants "sacrificed personal safety to better curate a positive social image". Their study is a design study: it proposes mechanisms (private conflict bubbles, system-attributed interventions, intent detection from repeated behaviour) but does not test them in live use.

A controlled vignette comparison (Germany, 2026). A University of Bamberg team randomly assigned 195 people in Germany to watch one of three Roblox video vignettes of a mild personal-space invasion: no safety feature, a safety bubble, or avatar blocking. Both features raised trust, controllability, satisfaction and use intention; the bubble also raised perceived safety and reliability, and the authors judge it "the more effective solution due to their preventive nature". This is a positive result for proactive tools. Its limits, as the authors state them: a mild scenario on 2D video rather than embodied VR, a general-population sample in one country, and only two features. They also note that a bubble may "deny meaningful interactions", and that users may switch it off to take part more fully. It did not test welcome approaches, consent, or disabled players.

A tested prototype (Puffer, 2025). A University of Washington team built Puffer, a social VR prototype with user-initiated personal bubbles, "preference badges" that state a player's norms, and prompts for bystanders to step in. In paired, simulated scenarios with 16 participants it found "evidence that Puffer can help protect players during emergencies, foster prosocial norms, and create more positive social interactions." Its own limits: participants were mostly aged 18 to 25 and from one US city; the scenarios "cannot replicate a realistic social VR gaming experience"; and its bubbles are still switched on by the player, the step the disabled participants above said carries a social cost. The paper does not report whether any participant was disabled.

What would unlock progress

Protection that is proactive but graded, and that the system carries rather than the person being targeted. The co-design study proposes using distance as a scale rather than a single cut-off: norm-setting at public distance, just-in-time intervention at social distance, and stronger support at personal distance (participants named automatic blocking and muting), with a consent step before anyone interacts with a disability signifier such as a virtual cane. It also proposes that interventions be attributed to the platform, not to the user who triggered them, so that using protection does not single anyone out. The Glasgow survey points the same way: VR-specific tools need the speed and familiarity of mute and block. Puffer shows that player-switched proactive tools can work in a staged session, and the 2026 vignette study shows that a bubble raises perceived safety and willingness to use. Consent as a design principle for social VR has been worked out on paper before: a 2023 interview study with 39 social VR users set out principles for "consent mechanics", a 2023 study ran design workshops with 18 potential users of dating spaces in VR, and a 2023 framework paper lists "assessing consent mechanics through user studies" among its steps without reporting such a study. None of the three built or tested a working prototype. No study found has tested a system-attributed, consent-based design in use, or reports testing any proactive design in use with disabled players or in ordinary multiplayer play.

How we’d know it’s being solved

  • In a scripted session with actors playing the intruder, the share of unwanted-contact attempts stopped before contact, set against the share of welcome approaches (consented conversation, game invitations) that still get through.
  • Time and number of actions needed to invoke protection, compared with mute and block (the speed users praised).
  • Whether participants, including disabled participants who show disability on their avatars, would use the tool in public, measured with a short social-cost rating ("would using this make others see me differently?").
  • In platform data, if any is ever opened: uptake of VR-specific tools against the 16–25% in the Glasgow survey, and repeat contact from blocked users.
  • A proxy to avoid: counts of blocks and reports. Rising use of reactive tools can mean more harm, not better protection.

Where to start

Facility-free door. Repeat and extend the Glasgow platform review from public sources: the platforms' safety pages, release notes and tutorials, and the free versions of the apps where a team member already owns a headset. For each of 10 to 12 social VR platforms, record which protections exist, which are on by default, how many steps each takes to reach, and whether a protection is visible to others when used. Then write a one-page design specification for a graded, platform-attributed protection, drawing on the co-design study's proposals. No partner and no data beyond public pages are needed. Skills: interaction design, systematic review, information design.

Build door (makerspace or lab with standalone VR headsets). Build a small multiplayer room with a free VR development kit, with two settings: the current boundary (on for strangers, off for friends), and a graded, consent-based version of the team's design. Test it with student volunteers, including disabled students reached through the campus disability services office, in supervised in-lab sessions. A trained team member plays a scripted intruder; there is no live harassment and no slurs, and the session uses only seated play. Measure the outcomes under "How We'd Know". What exists already: Personal Boundary (Meta), friends-only and trust-based filters (some platforms), mute and block everywhere. Consent mechanics for social VR have been proposed in interview and workshop studies (2023) but not built. The closest research prototype is Puffer (2025), whose bubbles and badges are switched on by the player; the closest controlled comparison is the 2026 Bamberg vignette study (bubble vs blocking vs nothing, on video). The team's version differs by letting a welcome approach through after consent, by attributing the intervention to the system rather than the player, and by testing with disabled players, whom the Puffer paper does not report recruiting. Access line: needs headsets (a campus makerspace may lend them) and human-subjects ethics review, because participants are exposed to simulated unwanted contact; the protocol should include stopping rules and a debrief.

Design-the-study door. Write a protocol a platform could run: a randomised comparison of default settings on live servers, with harm reports, session length and repeat contact as outcomes, and a pre-registered analysis plan. The team designs it and does not run it; running it needs a platform partner.

Genes

Related problems

Stuck the same way, in other fields.

field: water & energydilute, dispersed, and too potent to ignoreHundreds of thousands of engines, flares and coal mine shafts emit dilute methane, where conventional solutions often fall shortbuilt without the people who'd live with itsource: primaryfield: health & medicinethe planner needs a partnerSweden prescribes clocks and digital planners to people with dementia, but in Swedish interview studies they did not make time management independent: a partner still sets them up and prompts their use, and people without a partner's support risk missing outbuilt without the people who'd live with itsource: primaryfield: health & medicinethe light is on, the resident is in bedNursing homes are buying ceiling "biodynamic" lighting for residents with dementia, but a 12-month Dutch trial found no significant effect on sleep or daily rhythm while care routines may have kept many in bed; a US trial that put light where each resident sat did better, and the wearable sensors that show each resident's dose were often lost or removedbuilt without the people who'd live with itsource: primaryfield: health & medicinewhere to put the baby downIn Japan, some parents bathing a baby alone set the baby down on the washing machine or the bath lid; SG-standard bath-lid labels already warn against using a baby bath there, two babies died in neck floats in the bath after repeated warnings, and the appliance makers' 2026 response was a leaflet line, while parents cite narrow changing rooms and conveniencebuilt without the people who'd live with itsource: primary

Sources

  • Weerasinghe M, Macdonald S, Fiani C, O'Hagan J, Chollet M, McGill M, Khamis M. "Beyond Mute and Block: Adoption and Effectiveness of Safety Tools in Social VR, from Ubiquitous Harassment to Social Sculpting." IEEE Transactions on Visualization and Computer Graphics, 31(5), pp. 3275-3284, 2025, doi:10.1109/TVCG.2025.3549860 (accepted manuscript, University of Glasgow repository; published version, IEEE Xplore, read through library access), accessed 2026-11-17. Link ↗ Link 2 ↗
  • Zhang K, Killough D, Li XA, Yao Y, Zhao Y. "Safety vs. Social Image: Co-Designing Protection Mechanisms Against Ableist Harassment with People with Disabilities in Social Virtual Reality." In The 28th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS '26), October 25–28, 2026, doi:10.1145/3797867.3829041 (arXiv:2608.13532), accessed 2026-11-17. Link ↗
  • Li XA, Zhang K, Zhao Y, Yao Y. "Beyond Harassment: Exploring the Harm Experienced by People with Disabilities in Social Virtual Reality." In The 28th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS '26), October 25–28, 2026, doi:10.1145/3797867.3828993 (arXiv:2608.27390), accessed 2026-11-17. Link ↗
  • Liao Z, Zhao H, Kulkarni A, Chattrath SS, Zhang AX. "Building Proactive and Instant-Reactive Safety Designs to Address Harassment in Social Virtual Reality." Proceedings of the 28th ACM SIGCHI Conference on Computer-Supported Cooperative Work & Social Computing (CSCW '25), manuscript (arXiv:2504.05781), accessed 2026-11-17. Link ↗
  • Zheng Q, Do TN, Wang L, Huang Y. "Facing the Illusion and Reality of Safety in Social VR." CHI EA '22, Workshop on Novel Challenges of Safety, Security and Privacy in Extended Reality, 2022 (arXiv:2204.07121), accessed 2026-11-17. Link ↗
  • Xie R, Franke F, Birnstiel S, Morschheuser B. "Protecting Personal Space in Social Virtual Worlds: The Impact of Proximity-Control Features on Adoption Intention." 2026 IEEE Conference on Virtual Reality and 3D User Interfaces (VR): Proceedings, IEEE, pp. 410–419, doi:10.1109/VR67842.2026.00062 (accepted manuscript, University of Bamberg repository), accessed 2026-11-17. Link ↗
  • Meta, "Introducing a Personal Boundary for Horizon Worlds and Venues" (2022, with updates of 15 March 2022 and 13 December 2023), accessed 2026-11-17. Link ↗
  • Schulenberg K, Li L, Lancaster C, Zytko D, Freeman G. ""We Don't Want a Bird Cage, We Want Guardrails": Understanding & Designing for Preventing Interpersonal Harm in Social VR through the Lens of Consent." Proceedings of the ACM on Human-Computer Interaction 7(CSCW2):1-30, 2023, doi:10.1145/3610172 (full text from the US National Science Foundation's public access repository), accessed 2026-11-17. Link ↗
  • Zytko D, Chan J. "The Dating Metaverse: Why We Need to Design for Consent in Social VR." IEEE Transactions on Visualization and Computer Graphics 29(5):2489-2498, 2023, doi:10.1109/tvcg.2023.3247065 (full text from the same repository, record 10463395), accessed 2026-11-17. Link ↗
  • Homayouni C, Zytko D. "Consensual XR: A Consent-Based Design Framework for Mitigating Harassment and Harm Against Marginalized Users in Social VR and AR." 2023 IEEE International Symposium on Mixed and Augmented Reality Adjunct (ISMAR-Adjunct), pp. 360-364, doi:10.1109/ismar-adjunct60411.2023.00077, accessed 2026-11-17. Link ↗
  • VRChat, "Safety, Freedom, and the Future of VRChat" (19 December 2025), accessed 2026-11-17. Link ↗
  • VRChat, "Update: Age Assurance, Content Gating, and Safety Improvements in VRChat" (24 June 2026), accessed 2026-11-17. Link ↗
  • Modulate, "Rec Room's Proactive Approach to Reducing Toxicity" (vendor customer story), accessed 2026-11-17. Link ↗
Revision notes

This brief gets checked against its sources and contains the following revisions. Future revisions get added here.

Worth knowing

  • The evidence on disabled players comes from a design study with 11 participants and a survey of 67; no proactive design has been tested in live use.
  • The drops in repeat offences on one platform are the moderation vendor's own figures, with no stated method or independent evaluation.

Full check

Sources read

  • Weerasinghe et al., safety tools in social VR, survey of 100 players (IEEE TVCG, 2025): full text (published version and authors' copy, which agree on every figure used)
  • Zhang et al., co-designing protection against ableist harassment (ASSETS, 2026): full text
  • Li et al., harm experienced by disabled people in social VR (ASSETS, 2026): full text
  • Liao et al., the Puffer prototype (CSCW, 2025): full text
  • Zheng et al., safety in social VR, workshop paper (CHI, 2022): full text
  • Xie et al., proximity-control features and adoption intention (IEEE VR, 2026): full text (the authors' accepted manuscript)
  • Meta, Personal Boundary announcement and updates (2022–2023): full text
  • VRChat, two safety update posts (2025, 2026): full text
  • Modulate, Rec Room customer story: full text (a vendor's own account)

What changed

  • Not found in the sources, removed: in the build starting point, that many campus makerspaces lend headsets; it now says a makerspace may lend them.
  • Not found in the sources, replaced with what they say: that the prototype study recruited no disabled players; the paper gives no disability information, and the page now says so in both places.
  • Contradicted by the source, corrected: the uptake figures of 16–25% were described as proactive tools; one of them is a quick-reaction tool, so the page now says VR-specific tools, and a heading changed to match; that no study has tested a consent-based design with disabled players; the co-design study showed its participants a consent mock-up in a scripted scene, so the page now says none has been tested in use.
  • Links: addresses for the two VRChat posts and the vendor's Rec Room story, which carry claims in the main text, were added.

Left open

  • two findings credited to a 2023 study of four platforms [reported as the survey paper describes it; that study was not opened]
  • that human moderators addressed only 24% of incidents [cited from earlier work as the survey paper reports it]

Separate check of specific claims: Three 2023 studies that worked out consent principles for social VR on paper, without building anything, were added.

Earlier full check: Two findings were moved from a 2022 workshop paper that does not contain them to the 2023 study the survey credits, reporting is now 'among the lowest-rated' tools, and the co-design study's proposals and why disabled players show their disability on their avatars are now given as the papers put them.

Earlier full check: Both titles were narrowed because Meta's boundary is on by default only for strangers (display title 'A Bubble That Keeps Everyone Out' → 'A Bubble That Keeps Every Stranger Out'), a line credited to participants was given back to the authors, and a 2026 study with positive results for the bubble was added.