Three modes, one shell
An AI security platform for mid‑level content creators, detecting, alerting, and reporting content threats with a focus on large‑scale AI‑driven attacks.
Overview
Educational project from “Master UX Design for AI” (Maven, Rupa Chaturvedi).
SAFE is an AI‑powered security platform for mid‑level content creators (10K–500K followers). It detects, alerts, and reports content threats, with a focus on large‑scale AI‑driven attacks.
This case study focuses on the conversational interface: three modes (ASK, CHECK, REPORT) that adapt to the creator’s context and emotional state, from live incidents to proactive checks.
Challenge
Content creators face a new class of attacks: their work is stolen, manipulated, or re‑uploaded by AI agents operating at a speed they can’t match.
Current tools are expensive, not built for autonomous AI attacks, and fragmented across platforms — leading to financial loss, burnout, and damaged trust. Many attacks are only discovered when followers report them.
The challenge: design a system creators can actually trust, that returns a sense of control, and guides them from confusion to clear action.
Approach
Core principle: design a trust‑based system that returns control to the user. The solution applies human‑centered AI principles (user‑centric, ethical, reliable, usable) across three touchpoints:
- Alert, reactive response when an attack is detected. The system calms the user, explains what happened, and guides immediate action.
- Check, proactive scan before publishing. The user initiates, the system validates risk, and recommends next steps.
- Report, structured flow for formal escalation. The system collects evidence, drafts reports from collected data, and tracks status across platforms.
Research
Researched AI‑driven threats and current creator behavior.
Mapping
Mapped pain points and defined a focused persona.
Principles
Applied human‑centered AI principles across three touchpoints — Alert, Check, Report.
Flows
Designed end‑to‑end flows with happy paths, error handling, disambiguation, and fallbacks.
Prototype
Built a conversational system of variable‑based patterns that adapt in Base44.
Key Design Decisions
Three modes, one shell
Analysts switch between three states: reacting to an attack, checking a file, and filing a report. I kept one UI shell and split it into ASK, CHECK, and REPORT — so each state gets only the actions and guidance it needs.
Transparency over a black box
Instead of hiding scans behind a black box, I exposed system activity in real time. This gives analysts a clear status and an audit trail of what the system did and when.
Variable‑based response cards
I moved from fixed templates to variable‑based response cards. Each response is assembled from modular cards that adapt tone, pacing, and detail to the user’s state, while keeping patterns and intent consistent — at the cost of higher upfront design complexity.
Hi [USER_NAME].
[ACTION_STATEMENT] — it will take [TIME_ESTIMATE] minutes.
I’m checking [CHECK_ITEM_1], [CHECK_ITEM_2], and [CHECK_ITEM_3].
Impact & outcomes
- A conversational AI with three modes (ASK, CHECK, REPORT) that adapts from live incidents to proactive checks.
- End‑to‑end flows across happy paths, errors, and fallbacks — so analysts always know what happens next.
- A three‑panel layout (data, canvas, conversation) that stays fixed while the assistant’s behavior adapts, keeping AI activity transparent under stress.
- A variable‑based card system that adjusts tone and detail to language and emotion, prototyped in Base44.
- Demonstrates designing AI systems that build trust through transparency and emotional awareness — so analysts can make clear, confident decisions.