A University of Southern California’s Viterbi School of Engineering study released Tuesday found that large language models (LLMs) showed “strong performance in general communication, receiving high ratings for overall quality, empathy and specificity,” but were flagged for “providing varying levels of unauthorized medical advice,” raising “significant” safety concerns. Commonly reported issues among all LLMs included overgeneralization, making unsubstantiated assumptions, and providing unnecessary feedback. Google’s Gemini 1.5 Pro was most frequently flagged for “lacking empathy or emotional attunement,” while Meta’s Llama 3.3 was most frequently flagged for providing “unauthorized” health advice. OpenAI’s GPT-4 was commonly flagged for offering “unconstructive feedback and showing little personalization or relevance to the patient’s specific situation.” The report noted patterns of responses being “unintentionally judgmental” and that when asked to grade their own performance, “AI judges were unreliable, as they consistently overestimated their own performance and missed safety risks that human experts easily identified.” A November 2025 study in JAMA Network Open found that 13.1% of U.S. adolescents and young adults aged 12 to 21 years said they use generative AI for mental health advice.