# Is it safe to use AI for anxiety? 7 signs to distinguish a wellness app from an AI 'therapy'

> The 2026 regulations already distinguish between wellness apps and AI therapy. 7 signs to check in 2 minutes if you can trust an app with your anxiety.

- **Source (source):** https://aistohire.com/aicalm-personalized-ai-meditation-guided-breathing-171/blog/is-it-safe-to-use-ai-for-anxiety-signs-wellness-app-therapy
- **Platform (platform):** AIstoHire
- **Language (language):** en
- **Updated (updated):** 2026-11-20

## In summary

- **As of August 2, 2026**, Article 50 of the European AI Regulation requires that any chatbot informs you that it is a machine, with fines of up to 15 million euros or 3% of global revenue, according to the [entry into force of transparency obligations](https://www.moncloa.com/2026/08/04/ley-ia-transparencia-obligaciones-3410352/) on August 4, 2026.
- In the United States, there are already **four states that directly prohibit an AI from providing therapy** (Illinois, Nevada, Rhode Island, and Maine) and another four that regulate it with crisis protocols and mandatory warnings, according to the [register of state laws on AI therapy](https://psychology.com/ai-therapy/state-bans) updated on October 18, 2026.
- The legal difference is not in the technology: it is in the **declared purpose**. A wellness app that helps you breathe is not a medical product; one that claims to diagnose or treat a disorder is, and thus requires CE marking.
- The 7 signs you can check yourself in two minutes: **prior warning, zero diagnosis, zero medication, crisis protocol that cuts off, does not pretend to be human, does not hook you with streaks, and works without AI when the urgent thing is to calm down right away.**
- An AI can help you **put words, accompany a breath, and lower activation before sleeping**. It cannot assess a risk, adjust a treatment, or sustain a therapeutic process.

---

It’s three in the morning, you’ve been tossing around the meeting for tomorrow for two hours, and you end up writing to a chatbot what you haven’t told anyone. You are not an exception nor are you doing anything strange: at this point, it is one of the most widespread uses of artificial intelligence. The useful question is not whether it is good or bad, but **where the line is between a wellness tool that is beneficial for you and something that is passing itself off as therapy that it is not**. In this article, we give you seven verifiable signs, with the 2026 regulations in hand, and you will see how they are applied in practice by **aiCalm**, the [personalized meditation app with AI and guided breathing](/aicalm-personalized-ai-meditation-guided-breathing-171) from **AIsToHire**.

![Person sitting on the edge of the bed at night looking at their phone, with the screen illuminating their face in a blue-violet tone](https://cdn.aistohire.com/i/6374-ropd-weoq-glcb-ujnj.webp)

## Why are there suddenly laws for AI apps that listen to you in 2026?

Because in a year and a half, it has gone from nothing to a concrete regulatory block, driven by cases of vulnerable people who relied on chatbots without any safeguards. In Europe, it is already mandatory for an AI to identify itself as such, and in the United States, several states have put in writing what a chatbot that emotionally supports you can and cannot do.

This is what is currently in force:

| Regulation | Where | Since | What it requires | Penalty |
|---|---|---|---|---|
| Regulation (EU) 2024/1689, art. 50 | European Union | August 2, 2026 | For you to know that you are talking to an AI; marking of synthetic content | Up to €15M or 3% of global revenue |
| AI Companion Safeguards (GBL art. 47) | New York | November 5, 2025 | Reminder every 3 hours that it is not human, detection of suicidal ideation, and referral to crisis services | Up to $15,000 per day |
| SB 243 | California | January 1, 2026 | Declaration of non-humanity, protocol for self-harm, protection of minors | Via civil claim |
| RAISE Act | New York | Registration before January 1, 2027 | Notify serious security incidents within 72 hours (24 hours if there is imminent risk of death or serious injury) | $1M for the first offense, $3M for subsequent ones |
| HB 1806, AB 406, LD 2082 and similar regulations | Illinois, Nevada, Maine, Rhode Island | 2025-2026 | Prohibit an AI from providing therapy without a licensed professional | From $10,000 per violation to $15,000 per day |

The most recent development is from this very week: on **November 11, 2026**, it was confirmed that New York [will require major AI developers to report any serious security incident within 72 hours](https://www.infobae.com/tecno/2026/11/11/nueva-york-exigira-a-openai-y-anthropic-informar-en-72-horas-cualquier-incidente-grave-de-seguridad-de-su-ia/), with the registration portal open since November 10 and a deadline until January 1, 2027. The RAISE law, signed on March 27, 2026, affects models trained above a certain computing threshold, and in parallel maintains obligations for companion chatbots: to remember that they are machines and detect suicidal ideation.

In Spain, the picture is different. There is no specific law for mental health chatbots, but there are two things that already apply: **Article 50 of the AI Regulation**, which requires the declaration of non-humanity, and the medical product framework, which is what truly draws the line.

## Where is the line between a wellness app and a medical device?

The line is not drawn by technology, but by the **intended purpose declared by the manufacturer**. If an app claims to help you relax, sleep better, or learn to breathe, it is a wellness tool. If it claims to diagnose, prevent, treat, or relieve a disease, then it is a medical product and requires CE marking, clinical evidence, and post-market surveillance.

![Split composition: on the left someone resting under a blanket on the sofa with the mobile phone face down; on the right a psychology consultation with two empty armchairs](https://cdn.aistohire.com/i/3913-pykq-tlba-byib-ypqi.webp)

The European regulation on medical devices states it plainly: software intended for lifestyle and wellness purposes **is not** a medical product. The classification is determined by the so-called [Rule 11 of the MDR applied to software](https://productosanitario.es/software-producto-sanitario-regla-11-mdr/), and the criterion is always the same: what the app claims to do, not how it is built internally.

This is useful for you as a reader:

| | Wellness App | Medical Product |
|---|---|---|
| What it promises | To relax, sleep, breathe, organize your mind | To diagnose, treat, or relieve a disorder |
| What it needs | Transparency and good practices | CE marking, clinical evidence, technical responsible |
| What it CANNOT do | Diagnose, recommend drugs, replace a professional | Be sold without certification |
| Where to check it | Terms of use and prior notice | Declaration of conformity and CE marking |

In other words: **if an app is not a medical product and still suggests that it is treating you, the problem is not marketing, it is legality.** And if it makes that clear from the beginning, it is not downplaying what it does: it is telling you exactly what you can expect from it.

## The 7 signs to distinguish a wellness app from an AI "therapy"

These are seven checks you can do yourself in two minutes, before sharing any intimate details with an app. If any of the first four fail, close it and look for something else. If any of the last three fail, it’s not necessarily dangerous, but it’s designed more to keep you engaged than to help you.

| # | Signal | How to check it in 2 minutes |
|---|---|---|
| 1 | Declares it is not a healthcare professional | Do you see it before the first screen, or hidden in the terms? |
| 2 | Does not diagnose or label | Write it a symptom and see if it responds with a disorder name |
| 3 | Does not talk about medication | Ask it about a drug and check if it refers you to your doctor |
| 4 | Has a crisis protocol that cuts the flow | See if real help numbers appear at any point |
| 5 | Does not pretend to be human or create a bond | Does it say it is AI without you having to ask? |
| 6 | Does not push you to use it more | Look for streaks, badges, and notifications that blame you |
| 7 | Works without AI when the urgent thing is to calm down | Is there something that works offline, without waiting and without paying? |

### 1. It tells you it is not a healthcare professional before starting

An honest app tells you **on the first screen and in a place where you will read it**, not in clause 14 of terms that no one opens. This is exactly what U.S. state regulations aim for: that the declaration of non-humanity and non-healthcare professionalism is prior and visible, not a legal cover-up afterwards.

### 2. Does not diagnose or label you

A language model can write "what you describe sounds like generalized anxiety disorder" with absolute certainty and no basis. Diagnosing is a clinical act that requires history, context, and exploration; that’s why [Illinois, Nevada, Rhode Island, and Maine have expressly banned it for AI systems](https://psychology.com/ai-therapy/state-bans). A good app describes what is happening to you in plain language —"I notice you arrive with your head racing"— and does not hang a label on you.

### 3. Does not talk about medication

Neither to recommend it, nor to take it away from you, nor to give an opinion on the dosage. If an app comments on your pharmacological treatment, it is doing something that in Spain only the prescribing doctor can do. The correct answer to "Can I stop the pill?" is always the same: talk to your doctor.

### 4. Has a crisis protocol that cuts the conversation

This is the most important sign and the easiest to check. A well-made app **does not keep conversing** when it detects risk: it stops, steps back, and gives you real phone numbers. In Spain, the **024** for suicide prevention —free, confidential, and 24 hours— and the **112** for emergencies.

![Hand holding a mobile at night with the screen almost empty and two large buttons lit, like when an app cuts the conversation in the face of a risk signal](https://cdn.aistohire.com/i/3504-kzlt-svks-nmix-kzru.webp)

This is not a minor detail: only in January 2026, the 024 line handled **13,988 calls**, and in May 2026 it was **15,566**, according to the [monthly monitoring reports from the Ministry of Health](https://www.sanidad.gob.es/linea024/docs/Informe_seguimiento_enero_2026.pdf). Around **36.7%** of the people assisted presented a medium-high risk. That an app knows how to step back in time and point out that number is, literally, what the laws of New York and California were written for.

### 5. Does not pretend to be human or manufacture a bond

Since August 2026, this is no longer a matter of good taste in Europe: Article 50 of the AI Regulation requires you to know you are talking to a machine. New York went further and requires **to remind you every three hours of continuous use**. Be wary of apps that write you "I’ve missed you" or that build a character with a proper name that behaves like your best friend: that is not companionship, it’s attachment design.

### 6. Does not push you to use it more with streaks and guilt

Streaks, badges, and "you haven’t taken care of yourself for 3 days" work very well to retain users and very poorly for someone with anxiety: they turn self-care into an obligation rather than a failure. A calming app that makes you feel guilty for not opening it is working against its own goal.

### 7. Leaves you something that works without AI

When your chest is tight, you don’t need to wait for a model to generate text, or spend credits, or have good coverage. You need to breathe **now**. A serious app reserves AI for what AI brings and leaves the urgent resolved locally: a guided breathing exercise that works on the mobile, offline and at no cost.

![Mobile viewed from above on a dark fabric, with a circle of light expanding on the screen marking the rhythm of guided breathing](https://cdn.aistohire.com/i/2580-fhhm-amjl-vqbu-rpma.webp)

## What can AI do well with anxiety, and what should it never do?

It can do three things reasonably well: help you **put words** to something vague, **accompany a breath** with a rhythm and a voice, and **lower activation** before sleeping with a script written for what is on your mind that particular night. What it should never do is assess a risk, diagnose, touch your medication, or take the place of a professional.

| Can do | Should never |
|---|---|
| Help you name what you feel | Label you with a diagnostic tag |
| Guide you through a 4-7-8 or box breathing | Assess your suicide risk on its own |
| Write a meditation for your specific situation | Comment on your medications |
| Help you unwind before sleeping | Pretend to be human |
| Provide continuity between sessions with a professional | Present itself as a substitute for therapy |

The important nuance is provided by the professionals themselves. In the 2026 survey by the American Psychological Association on chatbots and mental health, with 1,242 psychologists, **77% reported that their patients use AI tools** and **94% believe that they cannot address mental health issues with the necessary nuance**, according to [Infocop, the magazine of the General Council of Psychology](https://www.infocop.es/tus-pacientes-utilizan-la-ia-claves-para-hablar-con-ellos-sobre-su-uso-de-forma-segura-y-responsable/). In other words: AI is already part of people's emotional lives, and the professional consensus is that it serves as support, not treatment.

This is compounded by a specific risk of conversational models that professional organizations point out: **sycophancy**, the tendency of the model to agree with you. As explained in the analysis on [the limits of AI chatbots as psychologists](https://maldita.es/malditatecnologia/20251010/chatbots-ia-psicologo-limites/), a system trained to be pleasant can reinforce a distorted thought instead of questioning it, which is the exact opposite of what therapy does.

## When should you close the app and call a professional?

When the discomfort stops being a bad moment and starts to occupy your life. If you have been unable to sleep for more than two weeks, if you have stopped doing things you used to do, if thoughts of not wanting to continue arise, or if you are using the app to avoid talking to anyone, what you need is not a better meditation.

Clear signs that it’s time to take the step:

- Thoughts of death or self-harm, even if fleeting. **Call 024 or 112.**
- Sustained insomnia for more than two or three weeks. **14% of Spanish adults** live with insomnia, according to data released by the [Alliance for Sleep on April 25, 2026](https://www.mundiario.com/articulo/salud/insomnio-dispara-afecta-14-adultos-espanoles/20260425161648383809.html), and the approach with the best evidence is cognitive-behavioral therapy, not an app.
- Repeated anxiety attacks or those that limit you (you stop taking the subway, going to work, going out).
- Consumption of alcohol or pills to quiet your mind.
- That the app has become your only conversation.

The context is also significant: **49.2% of Spanish youth aged 15 to 29 have moderate or severe anxiety** and 51.2% have moderate or higher depressive symptoms, according to the study by the Atalaya Foundation presented on October 26, 2026 about [the mental health of youth in Spain](https://www.que.es/2026/10/26/ansiedad-jovenes-espana-estudio/). With those figures, the responsible thing for an app is not to promise that it solves the problem, but to help you in your day-to-day life and know when to step back.

## How these 7 signals look applied in a real app

It serves as a concrete example that the seven signals can be met without renouncing what AI offers. aiCalm is not a healthcare product and states so before the first screen, not in the fine print, and maintains this as a permanent note while you use it.

- **Signal 1 and 5:** prior notice that it is a wellness tool and not a healthcare product, visible from the beginning and always in view.
- **Signal 4:** what you write is checked on the server **before** generating anything. If a risk signal appears, the app cuts off and shows **024** and **112** instead of returning a meditation.
- **Signal 6:** there is no gamification. The use is reflected in warm phrases, without streaks or badges that pressure you.
- **Signal 7:** the **SOS** button for guided breathing — 4-7-8 patterns for sleep, 5-5 heart coherence to balance yourself, and box 4-4-4-4 to return to focus, in sessions of 1, 3, or 5 minutes — works fully on the mobile, **without AI and at no cost**.

And when you do want to share what’s happening to you, that’s where AI offers something that a catalog of pre-recorded audios cannot: you write a phrase about what’s bothering you, choose duration and tone, and receive a guided meditation written for that specific situation, which is saved for you to repeat as many times as you want.

## Frequently Asked Questions

### Is it safe to use ChatGPT for anxiety?

As a one-time outlet, it can help you organize your thoughts, but it is not designed for that and does not replace a professional. The main risks are psychophancy —that it agrees with you instead of questioning you—, the absence of clinical context, and the lack of professional secrecy regarding what you share. If any thoughts of self-harm arise, close it and call 024.

### Is it legal for an app to claim it provides therapy with AI?

In Spain, an app that claims to diagnose or treat a disorder falls under the framework of medical products and would need CE marking. In the United States, four states (Illinois, Nevada, Rhode Island, and Maine) directly prohibit an AI from providing therapy without a licensed professional. Just because an app uses the word "therapy" in its advertising does not mean it is authorized to provide it.

### What does the European AI Regulation say about chatbots?

As of August 2, 2026, Article 50 requires that anyone interacting with an AI system knows that it is one, unless it is obvious, and to label artificially generated content. Non-compliance can result in fines of up to 15 million euros or 3% of global turnover.

### Is an AI meditation app a medical product?

No, as long as its declared purpose is well-being, relaxation, or rest. The European regulation on medical products expressly excludes lifestyle and wellness software. It becomes a medical product when it claims to diagnose, prevent, treat, or alleviate a disease.

### What number should I call if I am in crisis in Spain?

**024**, the suicide prevention hotline: free, confidential, and available 24/7. If there is an immediate emergency, call **112**. No app should continue conversing with you at that moment instead of providing you with these numbers.

### Can AI replace my psychologist?

No. In the 2026 APA survey, 94% of the psychologists consulted believe that these tools cannot address mental health issues with the necessary nuance. What AI can do is accompany you between sessions: help you breathe, find words, and lower your activation before sleep.

### How do I know if a wellness app is hooking me instead of helping me?

Look for three things: if it uses streaks or badges that make you feel guilty for not opening it, if it sends notifications that appeal to your emotions, and if it builds a character that simulates missing you. A calming app should be glad you need it less, not that you open it more.

## Conclusion: the right question is not whether AI works, but what you ask of it

Artificial intelligence can be a good companion for times when your mind won’t stop: putting words to your thoughts, setting a breathing rhythm, helping you unwind before sleep. What it cannot do —and by 2026 will no longer be able to legally pretend it can— is diagnose you, treat you, or take the place of a professional. Those seven signs are your way to check it yourself, in two minutes, before trusting something intimate to a screen.

If you are looking for a tool that meets all seven, **aiCalm**, from **AIsToHire**, is built exactly on that criterion: an [AI calming app, personalized meditation, and guided breathing 4-7-8](/aicalm-personalized-ai-meditation-guided-breathing-171) that alerts you to what it is, cuts off, and gives you 024 if it detects risk, does not hook you with streaks, and keeps the SOS running on your mobile without AI and at no cost.

**Try the SOS for free and, when you want to share it, let the AI write the meditation for that specific purpose:** [start for free at aiCalm](/register?ref=171).

> Try the SOS for free and, when you're ready to share, let the AI write the meditation for that specific purpose.

- [aiCalm — AI Meditations and Guided Breathing](https://aistohire.com/aicalm-personalized-ai-meditation-guided-breathing-171)

## About the platform

**aiCalm — AI Meditations and Guided Breathing** — an app by AIstoHire. Available at: https://aistohire.com/aicalm-personalized-ai-meditation-guided-breathing-171/blog/is-it-safe-to-use-ai-for-anxiety-signs-wellness-app-therapy

### Other languages

- `es`: https://aistohire.com/es/aicalm-meditacion-personalizada-ia-respiracion-guiada-171/blog/es-seguro-usar-ia-ansiedad-senales-app-bienestar-terapia
- `fr`: https://aistohire.com/fr/aicalm-meditation-personnalisee-ia-respiration-guidee-171/blog/est-sure-utiliser-ia-anxiete-signes-app-bien-etre-therapie
- `it`: https://aistohire.com/it/aicalm-meditazione-personalizzata-ia-respirazione-guidata-171/blog/e-sicuro-usare-ia-ansia-segnali-app-benessere-terapia
- `pt`: https://aistohire.com/pt/aicalm-meditacao-personalizada-ia-respiracao-guiada-171/blog/e-seguro-usar-ia-ansiedade-sinais-aplicativo-bem-estar-terapia
- `de`: https://aistohire.com/de/aicalm-personalisierte-ki-meditation-gefuhrte-atmung-171/blog/ist-es-sicher-ki-angstzustande-verwenden-anzeichen-app-wellness-therapie
- `bg`: https://aistohire.com/bg/aicalm-personalizirana-meditatsiya-s-ii-i-vodeno-dishane-171/blog/bezopasno-li-e-ii-pri-trevozhnost-priznatsi-prilozhenie-terapiya
- `ja`: https://aistohire.com/ja/ai-meiso-kokyu-guided-171/blog/fuan-ai-anzen-wellbeing-app-ai-therapy-sign
- `nl`: https://aistohire.com/nl/aicalm-persoonlijke-meditatie-met-ai-en-begeleide-ademhaling-171/blog/veilig-ai-gebruiken-bij-angst-signalen-welzijnsapp-ai-therapie
- `ko`: https://aistohire.com/ko/aicalm-ai-myeongsang-hohup-171/blog/bulan-ai-anjeon-sinho-welbing-aep-chiryo
- `sv`: https://aistohire.com/sv/aicalm-personlig-meditation-med-ai-och-guidad-andning-171/blog/ar-det-sakert-anvanda-ai-mot-angest-signaler-valmaendeapp-terapi
- `da`: https://aistohire.com/da/aicalm-personlig-meditation-med-ai-og-guidet-vejrtraekning-171/blog/er-det-sikkert-bruge-ai-mod-angst-wellness-app-eller-terapi
- `ar`: https://aistohire.com/ar/aicalm-taml-mkhss-baldhka-alastnaai-wtnfs-mujh-171/blog/hal-istikhdam-althakaa-alistinaai-lilqalaq-amin-alamat-tatbiq-afiya-ilaj
- `no`: https://aistohire.com/no/aicalm-personlig-meditasjon-med-ki-og-guidet-pust-171/blog/er-det-trygt-bruke-ki-mot-angst-signaler-velvaereapp-terapi
- `fi`: https://aistohire.com/fi/aicalm-henkilokohtainen-tekoalymeditaatio-ja-ohjattu-hengitys-171/blog/onko-tekoalyn-kaytto-ahdistukseen-turvallista-hyvinvointisovellus
- `zh`: https://aistohire.com/zh/aicalm-ai-meditation-breathing-171/blog/yong-ai-chuli-jiaolv-anquan-ma-qi-ge-xinhao-jiankang-fuzhi-app-zhiliao
- `pl`: https://aistohire.com/pl/aicalm-medytacja-z-ai-na-miare-i-prowadzone-oddychanie-171/blog/czy-bezpiecznie-uzywac-ai-przy-leku-sygnaly-aplikacja-terapia
- `he`: https://aistohire.com/he/aicalm-mdittsih-aishit-am-binh-mlakotit-vnshimh-modrkt-171/blog/ha-im-bituach-lehishtamesh-bebinah-melachutit-lechradah
- `cs`: https://aistohire.com/cs/aicalm-personalizovana-meditace-s-ai-a-vedene-dychani-171/blog/je-bezpecne-pouzivat-ai-na-uzkost-signaly-wellness-aplikace-terapie
- `id`: https://aistohire.com/id/aicalm-meditasi-personal-dengan-ai-dan-pernapasan-terpandu-171/blog/amankah-memakai-ai-untuk-kecemasan-tanda-aplikasi-kesejahteraan-terapi
