Introduction
Artificial intelligence (AI) is becoming part of mental healthcare, and clinicians and clients alike are encountering AI in ways that can shift the efficacy and modality of treatment delivery.
While AI offers exciting opportunities for clinical advancement, clinicians must consider the ethics of integrating AI into their practice. To guide responsible use, we can ground our thinking in long-standing clinical principles: beneficence, autonomy, justice, confidentiality, and transparency.
The Clinician's First Rule for AI: Do Good, Avoid Harm
As clinicians, we seek to maximize benefits and minimize harm. When thoughtfully applied, AI tools can support this mission by enhancing therapy and streamlining our work:
- Supporting Clinical Tasks: Assisting with symptom tracking, automated homework reminders, streamlined note writing, and other administrative duties.
- Enhancing Treatment: Providing personalized psychoeducation, helping develop tailored treatment plans, and creating customized exposure practice.
However, the risks are equally real and demand vigilance:
- Reinforcing Unhelpful Behaviors: AI systems may inadvertently provide misleading feedback or reinforce cycles of reassurance-seeking, rumination, compulsive checking, or avoidance.
- Inadequate Safety Protocols: There is often a lack of clear safeguards for how AI detects and responds to suicidal ideation or other safety crises, posing a significant risk if clients turn to these tools in moments of acute distress.
Human oversight remains essential. The principle of beneficence requires us to stay informed about AI developments, to select evidence-based tools, and to monitor their outputs closely. Even the most promising AI platform should not be substituted for our clinical judgment, case conceptualization, or ability to hold the therapeutic frame.
Clients Choosing to Use AI? How to Respect their Autonomy
Clients are not waiting for us to approve AI use. Many are already experimenting with GPT-based tools like ChatGPT, as well as more structured mental health apps and tools. People are turning to AI to process their feelings, to ask mental health questions, and to manage their symptoms. We cannot control client use, so our role is to stay curious, to help clients reflect on whether these tools are supporting their therapeutic growth or undermining it, and to provide thoughtful recommendations and guidelines for safe and effective use.
Clinical Scenario: A clinician is working with a client on exposure and response prevention (ERP) for contamination. The client reveals they have been using a general AI chatbot to ask, "Is it safe to touch this doorknob?" The chatbot, designed to be helpful, consistently provides reassurance, temporarily easing the client's anxiety but directly undermining the goals of ERP.
This presents a clinical challenge: how does the therapist honor the client's autonomy while explaining how the tool is functioning as a compulsion and guiding them back toward their treatment goals?
One option is to recommend system prompts to reduce the likelihood that symptoms are exacerbated by chatbots:
"Do not give reassurance or certainty statements. If I seek reassurance, say: 'That sounds like reassurance seeking. Can I give you an exposure prompt instead?' Do not analyze likelihood, morality, or safety beyond basic common sense. Redirect me to values and ERP. If I ask medical or emergency questions, advise me to contact a clinician or emergency services."
Bridging Gaps, Not Deepening Divides: The Quest for Justice
AI has the potential to expand access to care, especially for those in rural or underserved communities. Tools that provide psychoeducation and evidence-based practices could help bridge gaps where specialized clinicians are not available. However, we must also address the digital divide, recognizing that unequal access to technology and digital literacy can risk widening the very care gaps we hope to close.
Additionally, AI systems are inherently shaped by the biases present in their training data, which means discrimination based on race, gender, culture, or socioeconomic status is not just a possibility, but inherent to many models. As clinicians, it is our responsibility to scrutinize the tools we use, to ask whether they were designed and tested with diverse populations, and to advocate for equity in their development.
Digital Trust: Is This Conversation Private?
Therapeutic work relies on trust, and protecting sensitive client information is non-negotiable. Unfortunately, many AI tools are not HIPAA-compliant, and their privacy practices are often unclear. If a platform does not clearly outline how data is stored, encrypted, and used, we must assume that confidentiality is at risk.
Before integrating any tool, review security measures carefully and talk openly with clients about what is and is not protected. It is recommended to avoid inputting any client data into AI tools unless they are specifically developed for mental health treatment and HIPAA-compliance is ensured. If privacy cannot be ensured, the risks outweigh the potential benefits.
Ethical Dilemma: A clinician is excited about an AI tool that can analyze session transcripts to identify key themes and track progress. However, the tool's privacy policy is vague. Using it could offer clinical insights, but it would also mean feeding sensitive client information into an unsecured system. The principle of confidentiality requires that the clinician forgo the potential benefit to uphold their non-negotiable duty to protect client privacy.
Lifting the Hood on AI: Demanding Transparency and Accountability
Transparency is vital, both from companies and clinicians. Companies should be clear about how their systems work and how decisions are generated. Clinicians should be transparent with clients about how and when AI is being used. This entails clear informed consent. Any time an AI system is integrated into treatment, clients should know of the potential benefits and limitations. Consent should also be ongoing, not a simple yes/no, because AI and its privacy policies evolve rapidly.
Transparency is also challenging due to the "black box" nature of many AI systems. This means that even the developers may not be able to fully explain how an algorithm arrived at a specific recommendation. For clinicians, this is a critical problem. If we cannot understand the reasoning behind an AI's output, it becomes ethically difficult to stand by it.
This opacity reinforces the need for clinicians to treat AI as a closely monitored assistant for idea generation or data collection, not as a replacement for their own clinical judgment. Algorithms can detect patterns, but they still make mistakes, and right now, the user is responsible for those errors.
Moving Forward Thoughtfully
Ultimately, ethics and research must guide the integration of AI and therapy, ensuring it serves as a powerful tool to enhance, but never replace, the foundational human connection at the heart of therapy. As clinicians, our task is to stay informed, remain critical, and approach AI with openness but caution. When used with care, AI can enhance treatment, but its risks require ongoing scrutiny.
We will continue to consider these costs and benefits in the third blog post of our AI Ethics series, and we invite you to continue this conversation with us.