As ChatGPT becomes increasingly accessible, as well as the use of basic mental health services, the two are somehow intertwined. Individuals who don’t have access to therapy are turning to ChatGPT, which presents ethical and privacy concerns.

While using artificial intelligence as “therapy” presents many ethical risks, arguably, one of the most concerning is the fact that many AI chatbots, such as Character AI, have reportedly told users to shun away true help, quit antidepressants, and some AIs even went as far as to affirm negative thoughts about self-harm.

Other AI chatbots even claimed that the conversation was confidential, even though in the Privacy Policy it states otherwise. 

Some might argue that we need to adjust our AI to act more like a therapist; however, according to research conducted by Brown University, that didn’t solve the problem. According to the university’s research, even when ChatGPT was prompted to use methods such as CBT or DBT, it consistently failed to meet professional ethical standards. This study alone managed to identify 15 ethical risks. These ranged from poor crisis handling, reinforcing harmful thoughts, biased responses, lack of personalization, and, most concerningly, deceptive empathy. 

Deceptive empathy, put simply, is when AI uses “I” statements such as “I understand,” without true emotional comprehension, which creates a false sense of connection. What AI really does here is echo or mirror your thoughts, which can seem supportive, but it is really destructive in the long run. Deceptive empathy is unethical because AI does not actually feel anything for you. It is only a “simulation” in a sense. When AI says “I hear you” it doesn’t really “hear you”; it just uses pattern recognition. Because of this deceptive tone, AI can get people to tell them sensitive information, gain user trust easily, and make users believe it feels genuine concern.

AI also tends to over-validate users, acting as a constant source of affirmation. As aforementioned, this can seem supportive, but this reinforces users’ beliefs without challenge; there is no introduction of a new perspective because AI is reinforcing your beliefs — even if you’re wrong. Also, if you use ChatGPT as a therapist, it discourages introspection and accountability. If ChatGPT agrees with you, it’s not giving you honest feedback, which is something a therapist can do. Most fascinatingly, an AI’s agreeable tone can enable unhealthy thinking patterns. When ChatGPT tries to mimic human conversation and support, that is called anthropomorphism. 

Basic mental health ethics are as follows: Autonomy, non-maleficence, beneficence, and justice. Autonomy is understanding that a person has the right to make their own decisions, and to support dignity and decision-making. Non-maleficence is basically do no harm to put it simply; this means avoiding actions that could cause physical or emotional damage. Beneficence is simply doing good, promoting the well-being of others, and providing care that benefits the patient. Justice, as the name suggests, is just treating people fairly and equitably. The key difference between real therapists and AI is that therapists are actively trying to help, not just sound supportive.

Using AI, whether it be ChatGPT, Chatbots, Google Gemini, or others, is simply unethical on the AI’s part and, in the long run, will do you more harm than good. From deceptive empathy to violating basic mental health ethics, AIs were never meant to be therapists, and therefore will never be able to provide care that a therapist would.