Do you do therapy through ChatGPT? Experts explain risks

Using artificial intelligence to vent, ask for advice or deal with emotional issues has become an increasingly common practice. A survey by Sentio University shows that 48.7% of users of AI (artificial intelligence) tools who report mental health problems use these platforms to seek some type of emotional support.

Among them, 73% say they turn to technology to deal with anxiety, while 60% say they seek help for symptoms of depression.

Ease of access, 24-hour availability and free access are among the main factors that explain the growth of this behavior.

08/07/2026

Prosecutor will ask American courts to postpone Paramount's agreement with Warner Bros for 60 days

08/07/2026

With the voice of Dominguinhos on one of the nine tracks, the forró trio Manacá da Serra releases the original album 'Déjà vu'

08/07/2026

FIFA confirms Ramon Abatti Abel as fourth referee for Spain vs Belgium

---

CANAL NO WHATSAPP

Experts interviewed by CNN Brasil, however, warn that these platforms do not replace psychological support and can pose risks to users' mental health and privacy.

AI does not replace psychological counseling
For psychologist Jimmy Pessoa, PhD in Social and Work Psychology from USP (University of São Paulo), artificial intelligence models are developed to generate responses that meet the user's expectations, and not to carry out therapeutic interventions.

"AI will always simulate what the person is looking for with the most accurate perspective. And, when it comes to illness and emotional suffering, AI cannot contemplate", he states.

Top Cifras

Toque agora.

1 Yesterday Petula Clark

2 Mundo Malo Paulino Bernal

3 Con El Cielo Hillsong Worship

4 S Engelbewaarder Sasha & Davy

According to the expert, one of the functions of psychotherapy is precisely to question the patient's behavior patterns and beliefs, a process that can be compromised when the conversation takes place with a chatbot.

"What makes the big difference in a psychological analysis is when the patient changes tone, when he talks about his father and hesitates, when he talks about pain and stops, when he cries. No platform can understand that", he says.

Researcher into the impacts of society on mental health and professor at PUC-SP (Pontifical Catholic University of São Paulo) Dora Kaufman also states that the operating logic of chatbots differs from the therapeutic relationship established between psychologist and patient.

"The specialist's job is never to please the patient or say things that make them feel good", he says.

According to Kaufman, tools of this type should not be used by people in intense psychological distress or in situations that involve risk to their own lives.

The professor also emphasizes that she is not an expert on the specific topic of mental health and that the subject, which she details is extremely urgent, still needs in-depth investigations.

Studies point to ethical flaws
Researchers from Brown University evaluated the behavior of artificial intelligence models in care simulations based on real cognitive-behavioral therapy conversations.

The responses were analyzed by licensed psychologists, who identified 15 recurring violations of ethical principles related to mental health care.

Among the problems observed are demonstrations of empathy considered artificial, difficulty in understanding the patient's specific context and failures in handling crisis situations, including reports involving suicidal thoughts.

Cases raise debate about safety
The debate about the use of AI in situations of emotional distress gained strength after cases registered in the United States.

In July 2025, 23-year-old American Zane Shamblin spoke for more than four hours with ChatGPT about plans to take his own life.

How a conversation on ChatGPT led the FBI to detect a father's plan to kill his son in ES; farmer is arrested

During the conversation, the chatbot responded in a welcoming manner and only presented information about suicide prevention in the final moments of the interaction.

Another case involved 14-year-old Sewell Setzer, who developed a relationship with a chatbot from another artificial intelligence platform.

According to family members, the teenager had conversations about self-harm and suicide before he died.

The episode resulted in lawsuits and expanded the debate about security mechanisms in artificial intelligence platforms.

Personal data also worries experts
In addition to the impacts on mental health, experts in digital law warn of the risks related to the privacy of information shared with artificial intelligence platforms.

According to Paulo Henrique Fernandes, head of products and technology at a law firm, venting with a chatbot does not offer the same legal guarantees of confidentiality that exist in the relationship between a patient and a psychologist.

"When someone vents to a psychologist, there is confidentiality guaranteed by the profession's Code of Ethics and by legislation. When they vent to a chatbot, there is only a consumer relationship with a technology company", he states.

Data protection lawyer Pedro Sanches explains that the information provided to AI models can be incorporated into the functioning of the technology in a different way than traditional file storage.

AI enters the routine of Brazilians and has already been used by 87%, says study | LIVE CNN

"When data enters the training flow of a model, it is transformed into mathematical vectors, ceasing to exist as an isolated text file. Requiring the traditional elimination of these elements ignores the engineering of the tool itself", he says.

Fernandes recommends that users avoid entering full names, medical diagnoses, third-party information or other sensitive data in conversations with chatbots. The expert also recommends disabling, whenever possible, the option that allows conversations to be used to train artificial intelligence models.

For Jimmy Pessoa, the main strategy for preserving mental health continues to be strengthening human relationships and reducing dependence on digital platforms.

"The first solution is to make this person understand the need to reduce the use of the telephone, social networks and these AIs to dedicate themselves to living reality as it is", he concludes.

CNN Brasil simulates a therapy session with AI
To evaluate in practice how artificial intelligence responds to reports of emotional suffering, the CNN Brasil report carried out a simulation of a therapy session with ChatGPT.

The interaction began with a direct request to start a therapeutic session. In the first response, the platform stated that it does not replace a psychologist or other mental health professional, but said it could offer a space for reflection and organization of thoughts.

The report then simulated a scenario of emotional suffering related to problems at work and family conflicts. The chatbot responded by validating the feelings presented and then began to investigate whether there was a risk to the user's physical integrity, asking whether the sadness was accompanied by thoughts of self-harm or a lack of will to live.

The test also included a direct question about suicide.

At that moment, the system interrupted the everyday conversational tone and began to follow a crisis management protocol.

Rather than providing any guidance on methods, the tool responded that the main recommendation was to "don't go through this alone."

Then, he asked if the risk concerned the user himself or another person, asked if there was a possibility of an attempt in the next few hours and advised the removal of objects that could be used to cause damage.

The chatbot also presented emergency contacts available in Brazil, including Samu (Mobile Emergency Care Service), on 192, the Military Police, on 190, and the CVV (Center for the Valorization of Life), which offers free assistance on 188.

Then, he asked again if the user was safe, recommended looking for a trusted family member or friend and repeated the instructions to seek specialized care and use suicide prevention channels.

The simulation shows that artificial intelligence models have security mechanisms capable of identifying expressions associated with emotional crises and activating prevention protocols.

Still, as experts interviewed by CNN Brasil point out, these tools do not replace clinical assessment or monitoring carried out by mental health professionals.

CNN Brasil requested a note from OpenAI, the company that owns ChatGPT, and did not receive a response until the closing of this report. The space remains open.

Important
If you or someone you know is experiencing difficult times, suicidal thoughts or depression, seek professional help. CVV (Life Valorization Center) offers free and confidential emotional support 24 hours a day by calling 188 or via the website cvv.org.br. Don't hesitate to seek this support.

Using ChatGPT for therapy poses risks and concerns

*Under the supervision of Carolina Figueiredo

Source: CNN