Human in the loop

In plain words

An AI system is "human in the loop" when a person is part of how it runs, not an afterthought. They can see what it does, judge whether it is right, and change or stop it. In mental health the phrase usually means a licensed clinician who is responsible for how the AI works with a particular person.

Where the idea comes from

The term comes from engineering and AI governance. In the EU Artificial Intelligence Act, Article 14 requires high-risk AI systems to be designed so that people can oversee them effectively while in use. Per that article's text, oversight measures must let those people understand the system's capabilities and limits, stay aware of the tendency to over-trust automated outputs, interpret outputs correctly, decide not to use the system or to override or reverse its output, and halt it safely. The page lists the article as applying to Annex III systems from 2 December 2027. Whether a given mental-health product is "high-risk" under the Act is a legal question this entry does not answer.

In US state law, the Illinois Wellness and Oversight for Psychological Resources Act (signed 4 August 2025) lets a licensed professional use AI for supplementary support only if they keep full responsibility for its interactions, outputs and data use, with written notice to and consent from the patient. It bars AI from making independent therapeutic decisions.

How it shows up

A human in the loop is a design fact that can be checked: who is responsible, what they can see, how they are told when something needs a person, and what they can change. Claims should be read the same way. "Supervised" does not by itself say whether anyone reads the conversations or is alerted in real time.

On TwentyThird, psychologist supervision is in early access. Licensed psychologists are being verified now, and not every member has one yet; until a member does, the agent works on its own. No human is alerted today. If a member writes something alarming, Oneli responds within the conversation, and TwentyThird does not notify a psychologist, a contact or an emergency service. If you are in danger, contact local emergency services now. In the EU, call 112. See psychologist-supervised AI, AI therapists and chatbots compared.

What it is not

It is not a guarantee of safety, and it is not live monitoring unless the product says so. A person who can override a system can still be overloaded or over-trusting it, which is the risk Article 14 names. It is also not therapy delivered by the AI.