Dossier Pillar page

Ethics of AI in Mental Health

Your patients already use conversational AIs. Vendors offer you tools that are "ethical by design." This dossier gives clinical psychologists the bearings to evaluate these promises: the five principles of the AIEF framework, their limits documented by research, and an actionable evaluation grid.

A living dossier — last updated: July 2026 · Neither technophilia nor technophobia: an informed clinical practice.

Infographic of the dossier 'Ethics of AI in Mental Health': the five principles of the AIEF framework — beneficence, non-maleficence, autonomy, justice, explainability —, the practitioner's evaluation grid (intended use, expected benefits, potential risks, data and transparency, human oversight) and the enforceable benchmarks: code of professional ethics, GDPR, AI Act, HAS-CNIL guide. (Illustration in French.)
The dossier in one image: five principles to judge, one grid to evaluate, four enforceable benchmarks to decide. Click to enlarge. (Illustration in French.)
Infographic of the dossier 'Ethics of AI in Mental Health': the five principles of the AIEF framework — beneficence, non-maleficence, autonomy, justice, explainability —, the practitioner's evaluation grid (intended use, expected benefits, potential risks, data and transparency, human oversight) and the enforceable benchmarks: code of professional ethics, GDPR, AI Act, HAS-CNIL guide. (Illustration in French.)

Start here — the founding article

Five principles to judge AI in mental health: what the AIEF framework is worth, and what it lacks

More than 170 ethics frameworks published in ten years, zero demonstrated effect on patients. Why the principles nonetheless remain your best starting point — and how to convert them into a five-question evaluation grid.

Read the article (14 min)

The five principles, resource by resource

The framework of Floridi and Cowls (2019): the four bioethical principles that your professional ethics already operationalizes, plus a fifth specific to AI. Each resource: mechanisms, clinical case, points of caution and other perspectives.

1

Beneficence

Produce a demonstrated good, not a declared one

 2

Non-maleficence

Do no harm — including through slow, systemic harms

 3

Autonomy

Delegate knowingly, take back control at any time

 4

Justice

No two-tier mental medicine

 5

Explainability

The principle that makes the four others verifiable

The enforceable benchmarks

Unlike vendor charters, these frameworks are backed by obligations, governing bodies and sanctions. They are what ground your practice.

  • Your code of professional ethics — the foundation: the principles of AI frameworks restate its essentials, in a less binding form.
  • GDPR — mental health data benefits from reinforced protection that "well-being" applications often circumvent. See the resource Informed Consent and AI.
  • European AI Act (2024) — legal obligations for high-risk systems, including AI-based medical devices.
  • HAS-CNIL guide (February 2026) — the French sector-specific reference for the sound use of AI systems in care settings. Official PDF (French).

The key concepts of the dossier

Four resources complement the five principles — including the critical perspectives that put them in context.

AI Ethics Frameworks (AIEF)

170+ frameworks published, effectiveness never demonstrated: what "ethical AI" promises are really worth

Therapeutic misconception

When the user believes they are receiving care from a chatbot — the risk that disclaimers do not correct

Ethics of care

The care relationship at the center of ethical reflection, not abstract principles

Informed consent and AI

Consent in the face of algorithmic opacity and invisible data collection

→ All the resources in the Ethics category in the Key Concepts.

The rest of the dossier

This dossier is a living editorial project. Upcoming articles:

  • Beneficence put to the test: what "demonstrating a benefit" means
  • Therapeutic misconception: when the user believes they are in therapy
  • Beyond principles: care, Design Justice and virtue ethics
  • Digital CBT in three waves: an ethical reading

To follow the publications: the AI Watch or the media & training page.