Since the boom of Large Language Models (LLMs), countless apps, bots, and new companies have emerged claiming to use these tools to solve critical problems in mental health. This wave of new initiatives is understandable given the widely acknowledged inefficiencies in mental health care quality and accessibility as well as opportunities for improvement. However, despite over two years of these initiatives, there has been little to no marked improvement in global mental health, with many projects disappearing just as quickly as they emerged. We believe that this failure is not because LLMs or more general AI technologies have failed to ‘live up to the hype,’ or that any future versions of these models will succeed where their predecessors failed, but rather that the problems we seek to solve are inherently complex, multifaceted, and require such vast cross-disciplinary expertise that no one tool or set of tools can address them alone. While LLMs offer tremendous opportunities to accelerate mental health care and access, we must first rethink and construct a vehicle for mental health care that can be accelerated. In this presentation, I will share the guiding principles for the safe and responsible use of AI in digital mental health that we have developed at the Child Mind Institute through years of integrating AI tools in patient-, clinician-, and researcher-facing products. From risk assessment and crisis response in our digital journaling app, Mirror, to clinical report generation and the development of our accessible Symptom Checker, we have seen both the strengths and weaknesses of generative AI tools like LLMs. Learning from our failures, we have formulated a clear vision for how to use these tools to effectively improve the lives of our patients and clinicians. Our interdisciplinary team of engineers, clinicians, and community members with lived experience have distilled this vision into guiding principles that ensure the technologies we build are equitable and safe. If selected for this presentation, I will present these guiding principles, explain how we developed them, and provide concrete and actionable advice for how they can be implemented elsewhere. This talk aims to offer groups interested in applying AI tools to mental health work with a scaffold, while encouraging those already working in the space to reflect upon how these principles align with their current approaches. The application of AI to mental health is a rapidly evolving space, and the ideal outcome of this presentation is an audience engaged in a discussion about the guiding values we should embrace as we strive to increase access to quality care around the world.