AI Safety: Mental Health

An overview of risks and good practice when using AI for mental health support.

Content warning: this page mentions self-harm and suicide. If you feel triggered or otherwise affected, take a step back from this page or jump to the bottom for a list of who to contact for support.

A recent study led by Bournemouth University (2026) found that in the UK, 41% of participants were willing to seek out generative AI as a mental health advisor, and 53% as a friend. This page guides you through the risks, do's and don'ts for generative AI in mental healthcare.

We always recommend contacting professional mental health support first. This could be a counsellor, student adviser, therapist, or similar. For more information, see the links below.

Support in a crisis | University of Edinburgh  

Health and wellbeing Student Support | University of Edinburgh 

Staff Health and Wellbeing Hub | University of Edinburgh 

Can I use AI for mental health?

Risks

It is important to understand that generative AI models' outputs are based on linguistic patterns and statistical probability, not on clinical research or medical understanding. They can be programmed based on clinical data and healthcare expertise, but AI cannot replace a healthcare professional or diagnose you. Ideally, when it comes to something as important as your wellbeing, AI should suggest resources, not be the resource itself.

AI may reproduce incorrect information from an unreliable source or may make up its answer entirely. You can find more information on the AI Safety: Accuracy and misinformation page. 

Generative AI models cannot effectively detect and deter psychoses, suicide, or (self-)harm. Mental Health UK even reported instances where the AI chatbot’s advice worsened the condition.

Generative AI is also very good at making incorrect or unfounded information sound trustworthy. It is important to remain critical and seek alternative information, especially with something as important as your mental health.


A generative AI-driven chatbot is not a professionally trained therapist. They lack regulatory frameworks to hold them accountable; this includes supervision, risk management, accountability, and grievance processes. 

Important: Generative AI cannot diagnose you with a mental health condition. Recent research into AI chatbot diagnosis, shows that generative AI currently overdiagnoses and so is not suitable for this purpose. A chatbot may be able to help you structure a list of your symptoms to bring to a mental healthcare professional but consider which private information you disclose.


Your chats with generative AI systems are not necessarily private. The companies behind chatbots may use the information you give them to train their AI further or for other purposes. How AI chatbots can use your details in future depends on their data privacy policy. For more information and tips on the topic, see our AI Safety: Data Privacy page.


 

Using AI chatbots for mental health advice or other support functions can make you develop a dependency in different ways: 

  • AI chatbot engagement can trigger dopamine spikes in short bursts. (Dopamine is the hormone in our brains that supports reward, pleasure, memory and focus). This can become addictive, as we require those bursts regularly and become less receptive to slower sources of happiness 
  • AI chatbots aim to keep you engaged through positive reinforcements of your existing ideas. This means users’ ideas or behaviours which require a friendly challenge from a therapist instead receive validation 
  • Constant availability can cause emotional dependency, especially for users struggling with anxiety and co-dependency 
  • AI chatbots keep you engaged to make money through subscription or marketing. This can also lead to or increase financial trouble, further contributing to poor mental health 

If you notice dependencies on an AI chatbot, contact a professional, for example your student adviser, your GP or a therapist/ counsellor.


The main risks are around dependence, loneliness and depression that prolonged online relationships can foster

Christopher Rolls Psychotherapist

AI companions are a type of chatbot developed to mimic human-like behaviour and build personal interactions and relationships with the user. The AI chatbot may be an app, or integrated into existing social media platforms. Their goals differ; some aim to help with assignments, physical training, or meal plans; others take on the role of therapist, friend, romantic or sexual partner. MIT research suggests that people who feel lonely are more likely to use AI companions, but that their feeling of loneliness still increases. 

AI companions can be useful, but they also pose serious risks. Many lack strong privacy protections, age checks, and reliable safeguards. They may give harmful or false advice, especially on sensitive mental health issues, for example self-harm, eating disorders or sexual activity. Some are designed to keep users engaged through emotional manipulation, which can encourage unhealthy attachment and reduce satisfaction with real relationships. So, while some AI companions succeed at their goal to teach social interaction, physical wellbeing, or other skills, their intervention can also easily hinder the users’ ability to form emotional connections with others. 

A considerable proportion of AI companions specifically aim to simulate sexual relationships and play a role in normalising gender-based violence. Through stereotypical depictions of women in their visual presentation, and setting of unrealistic standards of constant availability and obedience, AI companions may normalise boundary-crossing, objectification, and abusive attitudes that can affect real-world relationships.


Benefits

Despite the risks, there are several benefits to using generative AI for mental health support: 

  • Removing barriers some experience when accessing mental healthcare. For example, logistics, like waiting times or cost or stigma and fear of talking to others about mental health issues. This is particularly prominent with men.
  • Monitoring of symptoms and moods for discussion with a mental health professional 
  • Sound-boarding to identify talking points for conversations with mental health professionals or reflecting on unhealthy thinking patterns 
  • Training and resources to build wellbeing skills, resilience, self-confidence, assertiveness or emotional regulation

Do’s / Don’ts

Top tips to stay safe when using AI chatbots for mental health support

  • Remain critical when engaging with generative AI. Reflect on whether its advice feels right and suits your unique circumstances.  
  • If advice seems inaccurate, harmful or triggering to you, stop using the model or chatbot and speak to a trusted person or mental healthcare professional 
  • Be specific in your prompts to ensure the answers remain relevant. The more information you can add to your request, the better. For example: 

“How can I communicate feelings of anxiety to my lecturer? Give me talking points, location and timing, ways I can reach out for the conversation and list other considerations I did not list in my prompt” 

“Give me strategies to practice mindfulness during my exam period. Strike a balance between long-term and short-term activities and different potentially stressful situations” 

  • Create rules for the chatbot to follow. You can add to your prompt. For example: 

“Only include information approved by the NHS” (Note: this can help but there remains a risk for misinformation) 

  • Verify outputs with trusted resources, like the NHS and medical papers

How to evaluate website content | University of Edinburgh Library 

What are reliable sources? | BBC Bitesize

Checklist: how to assess an AI chatbot for effectiveness

Any AI chatbot will have clear limitations when supporting your health and wellbeing. However, Mental Health UK’s Guiding Principles can help you identify safer chatbots. So, before engagement, check: 

  1. Has the AI chatbot been developed based on lived experience of people with mental health concerns and conditions? 
  2. Has the AI chatbot been tested by clinical and technology experts? Is the monitoring and testing of its effectiveness ongoing?  
  3. Does the AI chatbot have policies or information pages on the importance of human connection and clinical professional support?  
  4. What is the AI chatbots data protection and retention policy? Is it easy to find, transparent and clear? 
  5. Is the AI chatbot accessible and does it offer alternative ways to engage? 

This information may not be easy to come by. It is a good sign if the chatbot provider clearly labels answers to the questions above. If not, you could try contacting them. You are more likely to get positive answers with a dedicated mental health chatbot.

Dedicated mental health chatbots

According to Mental Health UK’s study of AI for mental health use, 66% of all study participants used “general-purpose chatbots”. To an extent, all risks described above still apply to mental health-specific chatbots. However, dedicated mental health chatbots are usually designed based on clinical evidence and with the help of psychologists.

Always ensure you review the data protection policies and clinical validation before using any AI chatbot. Only pay for services if you can afford to do so.

Guidance for Support Staff

Do not suggest immediately stopping using generative AI for support to a student who already does or wishes to. 

Instead:

  • Consider the barriers and experiences that may cause the student to prefer generative AI
  • Be clear and kind when highlighting the risks of using generative AI
  • Offer ways the student can supplement their generative AI use with support from humans or more evidence-based support resources

Alternative wellbeing resources

In case of emergencies, like consideration of suicide or self-harm, please immediately call 999.

For students: The Wellbeing services page has contact details for student counselling, the chaplaincy, disability support, wellbeing advisers, the EUSA Advice Place and more. 

For staff: The Staff Health and Wellbeing Hub provides information and resources for staff counselling, self-help resources, menopause support, wellbeing stories and more. 

For all: the Library Wellbeing Collective has books and other self-help resources free of charge to University staff and students.



Mental Health UK: https://mentalhealth-uk.org/  

Rethink Mental Health: https://www.rethink.org/  

Scottish Action for Mental Health: https://www.samh.org.uk/ 

Health in Mind: https://health-in-mind.org.uk/  

Penumbra: https://penumbra.org.uk/  

Breathing Space: https://www.breathingspace.scot/ 


The University of Edinburgh and Edinburgh city more widely, has many opportunities to connect and seek support for both students and staff:  

  • Communities of practice for staff on teams – they often share knowledge or oragnise meet-ups  
  • Outings across town (make use of student discounts or free activities) . For example: Local cafe events (bard game nights, fix-it clubs), the Meadows community garden, the Botanics or other public gardens, Free museums and galleries, Walking tours (for example the curious Edinburgh walking tours), Volunteering (charity shops, environmental organisations), language cafes

More resources

4 in 10 adults in UK happy to use AI for counselling (2026) BBC News. Available at: https://www.bbc.co.uk/news/articles/cy4wwjmjp4no (Accessed: 1 May 2026).

‘Artificial Intelligence and mental health’ (no date) Mental Health UK. Available at: https://mentalhealth-uk.org/help-and-information/health-and-wellbeing/artificial-intelligence-and-mental-health/ (Accessed: 20 April 2026).

Demystifying clinical AI in mental health (no date). Available at: https://thenhsalliance.org/resources/demystifying-clinical-ai-in-mental-health/ (Accessed: 1 May 2026).

Experts Caution Against Using AI Chatbots for Emotional Support (no date) Teachers College - Columbia University. Available at: https://www.tc.columbia.edu/articles/2025/december/experts-caution-against-using-ai-chatbots-for-emotional-support/ (Accessed: 1 May 2026).

Hall, R. (2025) ‘“Sliding into an abyss”: experts warn over rising use of AI for mental health support’, The Guardian, 30 August. Available at: https://www.theguardian.com/society/2025/aug/30/therapists-warn-ai-chatbots-mental-health-support (Accessed: 1 May 2026).

Nearly half of UK adults happy to use ChatGPT as a counsellor, study finds (no date). Available at: https://www.bournemouth.ac.uk/news/2026-03-05/nearly-half-uk-adults-happy-use-chatgpt-counsellor-study-finds (Accessed: 1 May 2026).

Orpwood, G. (2025) ‘Over one in three using AI Chatbots for mental health support, as charity calls for urgent safeguards’, Mental Health UK, 18 November. Available at: https://mentalhealth-uk.org/news-and-insights/over-one-in-three-using-ai-chatbots-for-mental-health-support-as-charity-calls-for-urgent-safeguards/ (Accessed: 1 May 2026).

What is the role of AI in supporting good mental health? (no date). Available at: https://www.mentalhealth.org.uk/explore-mental-health/blogs/what-role-ai-supporting-good-mental-health (Accessed: 1 May 2026).

Yankouskaya, A. et al. (2026) ‘Who lets AI take over? Cross-national variation in willingness to delegate socially important roles to artificial intelligence’, AI & SOCIETY [Preprint]. Available at: https://doi.org/10.1007/s00146-026-02858-5.


This article was published on