People are increasingly seeking mental health support from large language models (LLMs), but these AI-powered chatbots are often not equipped to recognize when a user is in crisis or to recommend escalation to a human specialist. Researchers at the Georgia Institute of Technology are working on projects aimed at making these interactions safer.
Three faculty members from Georgia Tech’s School of Interactive Computing—Professor Munmun De Choudhury, Associate Professor Rosa Arriaga, and Associate Professor Alan Ritter—have received 2025 Google Academic Research Awards to study artificial intelligence (AI) with a focus on trust, safety, and security. These grants were part of a broader initiative by Google supporting researchers nationwide.
De Choudhury and Arriaga will investigate potential harms LLMs may cause to people seeking mental health care. De Choudhury’s project, titled “Exiting Harmful Reliance: Identifying Crises & Care Escalation Needs,” is conducted in partnership with Angel Hsing-Chi Hwang from the University of Southern California. The team will review both real and synthetic chatbot transcripts alongside clinicians to identify language patterns that signal risk.
“A chatbot will always give a response and keep talking to you for however long you want,” De Choudhury said. “That may not be a good thing for someone in crisis. We need to know when the right response is to stop and suggest talking to a human.”
Arriaga’s research, “Dull, Dirty, Dangerous: Investigating Trust of Digital Resources Among Low-SES Mental Health Care Seekers,” examines how LLMs impact individuals with low socioeconomic status (SES). She has adapted terminology typically used for tasks suited for automation—dull, dirty, dangerous—to create a taxonomy of harms AI might pose in mental health contexts.
Arriaga aims to identify factors that lead low-SES users to trust chatbots for advice and how this differs across age groups and situations.
“We know one of the reasons some users go to LLMs is because they aren’t insured and can’t afford a therapist,” she said. “LLMs are available 24-7. Maybe it doesn’t start as a trust issue. Maybe it starts with availability.
“Some of these human-AI conversations that result in harmful mental health advice didn’t begin on the topic of mental health. In one case, the person started going to the machine for help with homework.
“Then this relationship evolved into personal matters. Should we constrain the system to limit itself to helping someone with their homework and not wander off that subject into mental health matters?”
Ritter’s project focuses on social media privacy tools using interactive AI agents designed to help users make informed decisions about what they share online. His work seeks ways for AI systems to assess risks in both text and images by identifying content that could reveal more than intended.
“We’ve been developing methods to assess risks in text, and now we’re extending that work to images,” Ritter said. “People post photos without realizing how easily they can be geolocated by advanced AI systems. A casual selfie near home might contain subtle cues about where you live, like a street sign, that reveal private details.”
The goal is for AI agents to review user posts before sharing them publicly—flagging risky elements and suggesting safer alternatives so users retain control over their privacy while maintaining freedom of expression.
Ritter plans on deploying advanced reasoning models capable of probabilistic privacy estimation so users can better understand whether text or images might inadvertently disclose personal information or location data before posting online.
Georgia Tech is recognized as a leading public research university focused on technology fields such as engineering, computing, sciences, business, design, and liberal arts according to its official website. The institution maintains its main campus in Midtown Atlanta and engages globally through partnerships, enrolling over 55,000 students annually while managing significant sponsored research funding as noted by school sources.
