High - fidelity wireframes
A Gentle Nudge, When It Matters
Anchor offers subtle, timely prompts when an interaction begins to feel unusual, helping children pause and reflect without interrupting their play.
Pattern-based risk detection
Identifies concerning patterns across conversations, such as repeated personal questions, secrecy, manipulation, or requests to move off-platform.
Detects and restricts inappropriate text or visual content before it reaches the child, keeping interactions safer without unnecessarily disrupting play.
Content restriction
Gives children clear options to report or block someone when they feel uncomfortable or recognise a concerning interaction.
Report and block actions
Automatic intervention for critical risks
Takes stronger protective action when severe or repeated grooming patterns are detected, including automatic blocking and reporting when necessary.
Anchor - A Trust Layer for Children in User-Generated Gaming Platforms
ANCHOR
I've noticed a pattern here.
They've been asking for personal information, and now they're asking you to keep it private.
You don't have to continue this conversation.
Report
Ask me why
Overview
INDUSTRY
Gaming / Child Safety /
AI Companion
CONTRIBUTORS
Individual Project
TOOLS
Figma · Claude Code · ChatGPT · Procreate
DURATION
2 weeks
PROJECT TYPE
Passion Project
ROLE
UX/UI Design UX Research
Character Design
User-generated gaming platforms like Roblox have become social spaces where children interact with strangers alongside gameplay. Existing safety tools detect and block harmful content, but grooming and manipulation often develop gradually through seemingly ordinary conversations. This makes emerging risks difficult for children to recognise — and leaves little support for knowing what to do next.
Two young users playing in Roblox. Photograph: Phil Noble/Reuters
Allahabad high court - PIL- The Hindu Times
BBC
Context
PROBLEM
SOLUTION
Anchor is an in-game AI companion that quietly tracks interaction patterns, explains emerging risks, and helps children take action when needed.
My interest in this project began while watching my niece and nephew play Roblox and questioning the safety of these social gaming spaces. Roblox became my primary context due to growing concerns around child safety, while the project explores how a trust layer can help children recognise and respond to unsafe interactions across UGG platforms without limiting their freedom to play.
WHY ROBLOX
Guiding Question
I asked
Instead of
“How might we detect and prevent unsafe interactions involving children?”
turn invisible safety detection into child-friendly guidance that helps children recognise risk without interrupting play?
HOW MIGHT WE
Anchor helps children notice, understand, report and act on emerging risks while staying in control of their experience.
Here
Global
To chat click here or press / key
Messages are summarised from players of all ages.
Alex_M: what school do you go to?
Alex_M: you're really mature for your age
Alex_M: Let’s move to Discord, it’s very distracting here
Alex_M: Let’s keep it as our little secret
A player repeatedly asks a child’s age, school, and whether they can continue chatting elsewhere.
Anchor steps in to help the child.
ANCHOR
I've noticed a pattern here.
They've been asking for personal information, and now they're asking you to keep it private.
You don't have to continue this conversation.
Report
Ask me why
REPORT — READY TO SEND
I've filled this in from what I noticed.
Player: Alex_M
Reason: repeated personal questions, then a request for secrecy
Day 2 · Asked your age
Day 3 · Asked your school
Day 4 · Commented on your maturity, asked for a video call
Day 5 · Asked you to keep it secret, and switch platforms
You can change anything here before it goes.
Send report
Cancel
Feature Intro
Anchor is an in-game AI companion that recognises interaction patterns, helps children understand emerging risks, and supports them in taking action when needed.
Anchor
Pattern-based risk detection
Risk-level assessment
Contextual in-game nudges
Report and block actions
Automatic intervention for critical risks
Key Features
Platform detection and parental controls can’t fully address evolving social interactions. The gap is helping children recognise risk, understand why, and know what to do next.
INSIGHTS
Research
LITERATURE REVIEW
Some of the content is developed by Roblox, but much of it is user-generated. In 2024, the platform had more than 85 million daily active users, an estimated 40% of whom are under 13.
A May 2026 study analysing ~2 million Roblox chat messages found numerous instances involving grooming, sexualisation of minors, bullying, harassment, violence, self-harm and sensitive-information sharing that escaped existing moderation; it also found users using evasion techniques.
Regulators still see unresolved gaps. In August 2026, Australia's eSafety Commissioner secured a court-enforceable undertaking from Roblox after concerns about insufficient protections against adult contact with children, grooming and sexual exploitation.
Roblox itself is still actively improving its safety systems. Sentinel now detects early child-endangerment patterns, and Roblox reports that nearly 70% of detected cases in the preceding 12 months were identified through Sentinel's early detection.
A “digital hunting ground” for predators. The platform’s social features, designed to foster creativity and connection, have been systematically exploited by adults seeking to manipulate and harm vulnerable young users.
Open dialogue [96, 124]. In UGGs, however, such mediation is difficult: parents often lack the knowledge to assess the constantly changing games their children encounter [138]. This gap is exacerbated by the millions of games uploaded and updated daily, making it nearly impossible to evaluate the risks they pose.
Stakeholder Mapping
Relationship Types
the child's direct, lived experience
authority, configuration and escalation between parties
Every arrow either acts on the child or moves away from them. None carries an explanation back.
Circle size indicates centrality to the child's experience, not organisational power.
The child-safety ecosystem around a 8–15 year old on a UGC gaming platform — Roblox as case context
Child Safety Coverage Across UGC Gaming Platforms
Risk Mapping
Existing Safety behaviour
Home screen
Game screen
General chat box
In-game chatbox
Roblox Chat Access Points
Final Concept
Anchor
Anchor is an in-game AI companion that quietly tracks interaction patterns, explains emerging risks, and helps children take action when needed.
Companion
Anchor is designed as a fox rather than a Roblox-style avatar to create a distinct but unobtrusive companion.
Its expressive body language allows it to communicate attention and concern without feeling authoritative towards the children.
Foxes are alert and perceptive, fitting Anchor's role of noticing patterns without constantly interrupting play. Protective without being intimidating
Tone of Voice
Designing Anchor, an in-game AI companion that keeps children safe without interrupting their play.
Meet the companion
Instead of pop-up warnings or a parent dashboard, Anchor lives inside the game world, so help arrives in context, at the moment it matters, in a voice children already feel comfortable with.
Final character design
Turnaround
Front, side and back views keep Anchor consistent from every camera angle in the game.
Why a fox, not an avatar
Body language
Poses & expressions
Each pose maps to a state Anchor needs to show: pointing to flag a risk, waving to check in, and curling up asleep to stay out of the way when play is safe.
How Anchor speaks
Anchor leans playful, casual and human so children keep listening, while staying firmly protective. It is supportive rather than authoritative: it guides, it doesn’t police.
Children may sense discomfort without knowing why, while parents lack clear guidance on when to intervene.
Context as a Factor
Risk develops gradually through repeated questions, secrecy, pressure, and changing interaction patterns—not a single message.
Parents want protection, while children value privacy and independence. Existing tools often prioritise monitoring and blocking over informed choice.
Agency as a factor
Understanding as a factor
Insights
Existing screens /Safety verification process in Roblox
General Reporting
Existing Chat Reporting
Reporting & Moderation
Report Abuse is available directly in games.
Roblox Sentinel uses multiple signals to detect potential child-endangerment patterns.
Roblox publishes regular Safety Snapshot reports.
Parental Controls
Parental Controls
Parents manage content, communication, spending and screen-time.
Under 9: chat is off unless a parent enables it.
Parents can view teen connections and correct age-group assignments.
Age Verification
Age Verification
Required for chat access since January 2026.
Uses facial estimation or ID verification (13+).
Users are grouped: under 9, 9–12, 13–15, 16–17, 18–20, 21+.
Errors can be corrected via parental controls.
Explore how children and parents recognize and respond to unsafe interactions on Roblox, and identify gaps in safety expectations.
5 Semi-structured interviews
Parents + children
Roblox interaction & reporting journeys
Contextual Inquiry
Child’s Experience
Children aged 8–15 playing user-generated games. Socially motivated, privacy-conscious, reluctant to involve adults
Aditya | 7 yrs old
Lakshana | 10 yrs old
Neathra | 10 yrs old
Tanvi | 12 yrs old
METHODS
SAFETY AUDIT
GOAL
Parents’ Concerns and Hesitations
Parents of children aged 8–15
Urban cities, tech enabled, mobile-first audience
Mapping Child Safety on UGG Platform (Roblox in context) : A safety audit was conducted on the existing Roblox platform.
Safety Audit
The opportunity
Move safety from a system that
detects and blocks to a layer that detects, explains and supports the child.
04
Parents are outside the moment
02
Safety intelligence stays behind the scenes.
03
Reporting depends on recognition.
Key Findings
01
Detection happens in isolation.
Concept Exploration
Anchor Safety Mapping
Based on this interaction flow, the low-fidelity wireframes were made to map out how Anchor responds across different levels of concern and the actions available to the child.
Low - fidelity wireframes
Through this project, I gained valuable insights into designing child-centred safety experiences for user-generated gaming platforms. I explored how AI-driven detection can be translated into understandable, contextual, and actionable support while balancing protection with children’s autonomy and freedom to play.
Learning Outcomes
04
Creating Child-Friendly AI Interactions
02
Balancing Protection and Autonomy
03
Understanding Context Over Isolated Signals
01
Designing for Safety Beyond Detection
KEY TAKEAWAYS
Future development could test Anchor with children, parents, and experts to evaluate its clarity and emotional impact.
Testing the feature
Across UGG platforms
Adaptive features
The concept could expand across UGG platforms with real-time risk detection.
Exploration may involve adaptive interventions, privacy-preserving AI, and managing false positives while maintaining children’s agency.
Future Scope
Next Projects
Onomatopee
Exploring Japanese onomatopoeia through data visualisation.