Artificial intelligence (AI) is increasingly being used for more than answering questions, producing documents, or finding information. For some, AI chatbots have become a source of companionship and emotional support. That development raises an interesting question for those working with adults who may lack capacity: how should capacity to make decisions about the use of an AI companion be assessed?
The Court of Protection has already considered capacity to use the internet and social media, and capacity to make decisions about contact, but an AI companion does not fit neatly into either category.
In this Keynote, mental capacity and Court of Protection specialist Zena Bolwig considers how the Mental Capacity Act 2005 might apply to AI companionship, including how the relevant decision and information should be identified.
The reality of AI companions
Research published by the UK AI Security Institute found that, in a representative survey of 2,028 UK participants, 33% had used AI for emotional purposes during the previous year, with 8% doing so weekly and 4% daily. There is also increasing evidence that some users experience their relationship with AI in genuinely relational terms. Research published in Nature Human Behaviour in September 2026 considered users’ reactions to significant changes to Replika and ChatGPT and found evidence of attachment-related feelings of loss and distress.
Of course, an intense, unconventional, or apparently unwise relationship with an AI would not, without more, provide a basis for concluding that somebody lacks capacity. The legal question would remain whether because of an impairment of, or disturbance in, the functioning of the mind or brain, the person is unable to make the particular decision in question.
Is this internet use or contact?
In 2019, the Court of Protection considered capacity to use the internet and social media separately from capacity regarding contact and care, and set out the sorts of information which might be relevant when assessing capacity in that domain. The Court of Appeal subsequently emphasised that relevant information is not a fixed checklist to be applied in the abstract; it must be identified by reference to the particular decision facing the particular person. The existing authorities on internet and social-media use may be helpful, but they do not provide a ready-made test for ‘capacity to use AI’, nor should a new generic checklist simply be created for that purpose.
Unlike WhatsApp or Facebook, where technology facilitates communication between people, with an AI companion the technology is itself the apparent participant. Some considerations familiar with contact cases, including difficulties understanding relationships, reciprocity or deception, may therefore be relevant, although the categorisation is not straightforward.
The difficulty is illustrated by A Local Authority v ZX [2026] EWCOP 3 (T3). In that case, the expert ultimately accepted that online messaging should, on the particular facts, be considered as part of contact, and described the difficulty in disentangling contact, sexual relations, and internet and social-media use.
What might the relevant decision be?
The Mental Capacity Act 2005 requires consideration of capacity in relation to the particular matter requiring a decision. Using AI to check train times is plainly different from using an AI chatbot for several hours each day as a trusted confidant, friend, or partner.
In an appropriate case, the question might be whether the person has capacity to decide whether and how to use AI for companionship or emotional support. Alternatively, where AI influences decisions about contact, care, or finances, capacity may need to be assessed in relation to those underlying decisions.
What information might be relevant?
Once the correct decision is identified, what information must the person be able to understand, retain, and use or weigh in order to make it? The answer will depend on the individual circumstances and the particular decision, but where the decision concerns AI companionship itself, it might be relevant, for example, whether the person understands that:
- the AI is not a human being and, notwithstanding its apparently personal responses, does not have human feelings or intentions;
- the system may generate responses which simulate affection, concern, agreement, or intimacy without possessing corresponding feelings, intentions, or commitment;
- its responses may be inaccurate, inappropriate, or influenced by the way in which the system is designed, trained, or altered; and
- what the AI says may influence decisions the person makes about their relationships, finances, health, care, or other areas of their life.
Privacy and the use of personal information may also be relevant.
A person may understand that an AI companion is not human while nevertheless finding the interaction emotionally significant. The question is not necessarily the strength of that attachment or reliance, but whether difficulties understanding or weighing the artificial nature of the interaction affect their ability to make the particular decision.
Where might the first cases arise?
The first reported cases may arise not from AI companionship itself, but because AI has influenced another decision. An AI companion encouraging somebody to stop seeing family members, transfer money, or refuse treatment may raise questions about capacity concerning contact, finances, or treatment, respectively. In other cases, AI use may itself require determination, particularly where restrictions on access to a companion system are being considered.
It will be important to distinguish between cases where AI has influenced a person’s decision and cases where that influence is relevant to their capacity to make it. People regularly make decisions based on advice from others, including advice which may be mistaken or unwise, but the position may be different where, for example, a person is unable to understand or weigh the nature or limitations of the AI’s responses, and that inability materially affects their decision about whether to maintain contact with family members or transfer money.
If a person is ultimately found to lack capacity, a separate best-interests question may then arise about the weight to be given to a relationship which, although artificial in origin, may have considerable emotional significance for them.
An emerging question for the Court of Protection
Existing MCA principles will apply to AI companionship, but their application will depend on the particular circumstances. The challenge for the Court of Protection will be distinguishing an unusual or unwise attachment from an inability to make the relevant decision.
If you have questions or concerns about mental capacity or cases before the Court of Protection, please contact Zena Bolwig.
For further information please contact:
