Sam Altman Warns: The Risks of Using ChatGPT as a Substitute for Real Therapy

Tech
By 24matins.uk,  published 29 July 2025 at 14h01, updated on 29 July 2025 at 14h01.
Tech

Sam Altman, CEO of OpenAI, has cautioned against using ChatGPT as a substitute for professional therapists. He emphasizes the importance of human expertise in mental health care and warns users not to rely solely on AI for psychological support.

Tl;dr

  • No legal protection for AI-user conversations.
  • Emotional use of ChatGPT remains very rare.
  • Future AI may challenge privacy boundaries.

When it comes to digital interactions, the issue of confidentiality with artificial intelligence has rapidly taken center stage. Unlike conversations with professionals such as doctors or lawyers, there is currently no legal framework safeguarding what users tell AI systems. This point was underscored recently when Sam Altman, CEO of OpenAI, raised public concerns about the security of personal discussions conducted via ChatGPT. During his appearance on « This Past Weekend w/ Theo Von », Altman reminded listeners that, for now, AI enjoys none of the protections enshrined in doctor-patient or attorney-client privilege.

The Reality Behind User Interactions

Despite these regulatory gaps, it appears that most people do not treat AI as a confidant. Joint research by OpenAI and the MIT found that « L’engagement émotionnel avec ChatGPT est rare dans les usages réels. » The vast majority of user conversations show no sign of empathy or emotional depth—an observation echoed by Anthropic, creator of the chatbot Claude. According to their data:

  • Only 2.9% of interactions had any emotional content.
  • Requests related to companionship or role-play: just 0.5%.

Clearly, therapeutic or affective uses remain on the fringes for now.

The situation becomes more complex when legal matters arise. If required by court order, OpenAI could be compelled to provide a user’s entire conversation history—even those messages thought to be deleted. A recent dispute with the New York Times, involving data retention and deletion practices, led a judge to instruct OpenAI in May to preserve « toutes les données d’historique qui auraient normalement été effacées », even against user wishes or existing privacy laws. Thus far, nothing guarantees users a level of secrecy comparable to established professional privilege.

Toward a New Frontier?

Nevertheless, as models evolve—especially with the anticipated release of GPT-5, promising more natural dialogue and improved memory—the landscape could change rapidly. Some may soon feel inclined to seek advice or solace from their virtual assistants. In this context, individuals will need to weigh carefully what they choose to share before the distinction between technical tool and trusted confidant fades away entirely.