Log in or create a free Rosenverse account to watch this video.
Log in Create free account100s of community videos are available to free members. Conference talks are generally available to Gold members.
When AI Becomes the User’s Point Person—and Point of Failure
Summary
Imagine slipping on a sleek pair of smart glasses. Not only do you look sharp, the glasses capture everything you see, hear, and do. Your AI assistant—built into the glasses and synced to your email, social media accounts, health apps, and finances—manages your life. It’s tasked with paying bills, booking trips, replying to messages, even helping you swipe right. Over time, you find yourself chitchatting with your AI assistant. You call him Charlie. Now imagine you’re a threat actor. That trust between user and AI assistant? It’s your entry point. If your product is powered by AI, you’re not just designing features—you’re designing an entire relationship. You’re designing Charlie. Let’s talk about where that goes wrong—and how to get it right.
Key Insights
-
•
Users often do not understand why AI-powered systems request extensive personal data, increasing privacy risks.
-
•
Trust in AI agents can become excessive, creating new vectors for manipulation by threat actors.
-
•
Security issues typically occur beneath the surface until alerts disrupt the user experience, often causing frustration.
-
•
Prompt injection attacks pose a novel threat where malicious inputs manipulate AI agents to access sensitive user data.
-
•
Multimodal AI interfaces introduce complexity in security decisions, increasing chances for user errors.
-
•
Secure by default settings reduce burden on users and improve overall protection without requiring user intervention.
-
•
Cross-disciplinary collaboration between UX, security, product, legal, and compliance teams is crucial for safer AI design.
-
•
Users need clear, contextual guidance during onboarding to make informed decisions about data sharing and security settings.
-
•
Transparency about AI limitations and giving users the option to reverse AI actions are essential for building trust.
-
•
Threat actors are likely to exploit growing AI access to personal data and automate vulnerabilities discovery.
Notable Quotes
"When a product is powered by AI, you're not just designing the features; you are designing an entire relationship."
"Charlie is like the most annoying coworker who constantly surfaces problems but never offers solutions to Alice."
"Threat actors probably know your system better than you do and are looking for any entry points to exploit."
"Alice often perceives Charlie as just another barrage of alerts filled with jargon she doesn't understand."
"Prompt injection attacks can trick AI agents into accessing private data like emails without the user realizing."
"People become incrementally more comfortable giving away data because they see the value AI provides."
"We need secure defaults that protect users out of the box without them having to figure it out."
"Alert fatigue is real; users can't be burdened with constant security decisions or they'll ignore them."
"Giving users the ability to reverse AI-driven actions is critical but currently underexplored."
"If Charlie has been tampered with, Alice needs a clear way to be alerted that she shouldn't trust it."
Or choose a question:
More Videos
"Our interview process has just three steps: recruiter screen, hiring manager screen, and a final panel presentation day."
Kurdin Bazaz Liz Rytting Alex KarrCulture, DIBS & Recruiting
June 10, 2021
"Design influence is messy and requires clear articulation of how we serve business and customer success."
Jack MoffettUX Metrics That Matter and The Future of our Design at Scale Conference: A Community Conversation
September 22, 2022
"We had to allow people to grieve first before we could make a plan for healing."
Kyria Stephens Marlon KernerPower to Heal: Civic Design in the Aftermath of Tragedy
November 17, 2022
"Design has to be fast, accessible, actionable, and intuitive, whereas science is often slow and hard to act on."
Lin NieWhen Thought-worlds Collide: Collaborating Between Research and Practice
March 10, 2021
"If a bot can’t handle the query, it goes to a human—it's like Mechanical Turk for AI."
Greg NudelmanDesigning Conversational Interfaces
November 14, 2019
"The most successful programs meet stakeholders at their highest point of motivation to participate in research."
Trisha TerharEmpathizing with the Empowered: Non-Researcher Responses to Democratization
March 10, 2022
"We kept feeling the same consistent pain around some truths for our design system."
Charles Lee Jennie YipBuilding a New Home for the Atlassian Design System
October 22, 2020
"Someone who thinks a study will run perfectly every time probably isn’t very experienced or realistic."
Steve Portigal Susan Simon-Daniels Tamara Hale Randolph Duke IIWar Stories LIVE! Q&A-Discussion
March 30, 2020
"Just because empathy isn’t always visible at the top doesn’t mean you can’t be the change."
Deanna Washington Bria Alexander Jon Fukuda Saara Kamppari-Miller Farid SabitovConnecting the Ops: Plenary Panel and Closing Circle
September 9, 2022
