WHO Ethical Principles for AI in Health: A Consumer Guide for Mental Wellness Apps
WHO Ethical Principles for AI in Health: A Consumer Guide for Mental Wellness Apps
International guidance can feel distant from your phone screen. Yet the World Health Organization's work on ethics and governance of AI for health is one of the clearest frameworks for deciding whether a mental wellness product respects you as a person or treats you as engagement fuel[^who].
Why WHO language still applies to non-medical apps
Many chatbots disclaim being medical devices. Even so, they collect sensitive disclosures, shape beliefs about mental illness, and may steer users toward or away from professional care. That is ethically weighty even when regulators classify the product lightly.
The six principles in consumer terms
Protect autonomy: You should know what the system is, what it cannot do, how to stop using it, and how to delete data without a runaround. Informed consent is not a one-time banner; it is an ongoing relationship.
Promote well-being, safety, and the public interest: Designers should meet reasonable standards for accuracy within scope, monitor harms in the wild, and avoid optimizing solely for time-on-app when that competes with sleep or offline relationships.
Transparency and explainability: You should know when AI is speaking, what inputs influence outputs at a high level, and when answers are uncertain. Mental health is too high-stakes for mystery boxes.
Accountability and human oversight: Someone at the organization should be responsible for safety updates, content policies, and redress when things go wrong. "The model decided" is not an acceptable end state.
Inclusiveness and equity: Training data and product defaults should not systematically disadvantage dialects, disabilities, or lower-income users who cannot pay for premium tiers that unlock basic safety.
Sustainability: Teams should plan long-term maintenance, environmental costs of huge models, and fair labor for the humans who label safety data.
How to score a vendor in ten minutes
Open the privacy policy and safety page. If you cannot find crisis guidance, if training data language is vague, or if accountability contacts are missing, treat that seriously. Compare what you read with WHO's principles as a moral checklist[^who].
Reflektion alignment
Reflektion presents itself as non-clinical reflection support. We still believe transparency, safety, and respect for autonomy matter because vulnerability is involved whenever people speak about their inner lives.
Equity in practice, not only slogans
Inclusiveness means testing with diverse dialects, literacy levels, and assistive technology users. It also means pricing that does not gatekeep safety features behind ultra-premium tiers, and content moderation that protects LGBTQ users from conversion tropes or harassment templates accidentally reinforced by models trained on biased web text.
Accountability in small teams versus giants
Startups may lack the compliance army of hospital vendors, but they can still publish named safety owners, publish update changelogs when model policies shift, and respond promptly to user reports. Transparency scales with culture more than headcount.
When principles collide
Autonomy versus safety tension appears when users demand unfiltered answers. Ethical products refuse certain requests while explaining why, instead of silently jailbreaking. Document those tradeoffs for your community so trust can survive disagreement.
[^who]: WHO publication: Ethics and governance of artificial intelligence for health.