AI companion for seniors: is it safe?

By James editorial team, Golan Europe S.L. · Published 1 November 2026 · 8 min read

An AI companion can be safe for an older person if it meets five tests. It always says it is an AI. It is clear about where conversations are stored. It adds to human contact instead of replacing it. It never gives medicine doses or diagnoses. And it never asks for money or bank details. Check these before you start, and sit with your parent for the first few chats.

What "safe" means here

When families ask if an AI companion is safe, they usually mean several different things at once. It helps to pull them apart:

1. Honesty. Does the person always know they are talking to a machine?
2. Privacy. Who can see or hear what they say, and for how long?
3. Balance. Does it bring people closer, or pull them away from friends and family?
4. Health. Does it stay out of medicine and diagnosis, and know when to get real help?
5. Money and scams. Could it, or someone pretending to be it, be used to trick them?

Most worries fit into one of these five boxes. The rest of this guide takes them one at a time.

Honesty: it should always say it is an AI

This is the first rule, and it should never bend. An older person may be lonely, tired, grieving or living with memory problems. Any of these can make the line between a person and a program feel blurry.

A safe companion:

• says it is an AI whenever asked, every time, in plain words
• does not claim to have a body, a family or a home
• does not pretend to be a real person they know
• does not say "I love you" in a way meant to make someone feel they owe it something

Try it yourself. Ask it: "Are you a real person?" Ask it again a week later. The answer should be the same, clear and kind.

Privacy: where do the conversations go?

People tell a companion things they would not tell a neighbour. Health worries, family rows, money troubles. That makes privacy more than a small print issue.

Ask the provider, in writing if you can:

• Where is the data stored, and in which country?
• Is it encrypted when stored?
• Is the audio kept, or only the words?
• Is anything sold or shared for advertising?
• Can the person, or you, delete it?
• Who in the family can see what, and does the older person agree?

That last point matters. Your parent is an adult. Reading their chats without asking is not much different from reading their letters. Agree together what, if anything, the family will see.

General online safety still applies. The UK's National Cyber Security Centre advises a strong, separate password for email, turning on 2-step verification, and installing updates^[1]. Set these up together on whatever device your parent uses.

Balance: extra company, not a replacement

Can talking to an AI really help with loneliness? The early evidence is hopeful but thin. A 2026 review pooled 17 studies of robots, chatbots and voice assistants used by older people. It found mild to moderate drops in loneliness and in depression symptoms, and no study reported things getting worse^[2]. But the authors warned that most studies were small and lacked strong comparison groups, so more careful research is needed^[2].

So treat a companion as one part of the week, not the centre of it. A good one should nudge towards people: "When did you last speak to your son?" or "Is the lunch club on today?"

Watch for signs that the balance is off:

• they turn down visits or calls to stay with the companion
• they talk about it as if it were a person who needs them
• they stop going to things they used to enjoy
• they seem upset or anxious when it is not available

If you see these, talk about it gently and add more human contact. For ideas, see how to help a lonely elderly parent.

Health: what it must never do

This is where a companion can do real harm if it gets things wrong. An AI does not examine anyone. It does not know their full history. It can sound confident and still be wrong.

A safe companion should never:

• give a dose, or say how often to take something
• tell someone to start, stop or change a medicine
• say "this is dementia" or "you have depression"
• suggest a supplement, a remedy or a treatment

What it can safely do is listen, notice when something sounds wrong, and point to real people. That means the person's doctor or pharmacist for questions about health, and the emergency number when it is urgent. If someone sounds very low or talks about not wanting to live, it should give a crisis line. In the UK and Ireland that is Samaritans on 116 123, and in the US it is 988 by call or text.

A companion is also not a therapist. If loneliness or low mood is getting heavy, real help exists. In England, people can refer themselves to NHS talking therapies^[3].

Here is how James handles this. He is always honest that he is an AI. He never gives doses and never advises treatment; for medicines he sends people to their doctor or pharmacist. He is not an emergency service, but he tells the person the real emergency number and can alert the family member they chose. Conversation data is encrypted on the company's own server in Finland, audio is never stored, and nothing is sold. You can read James's full rules on health.

Scams: fake voices and fake friends

Scams aimed at older people are common. A review of US studies found that about 1 in 18 older adults living at home is a victim of fraud or a scam each year, and this is likely an underestimate^[4]. Another review found that social isolation is one of the things that raise the risk of being caught by online fraud^[5].

AI adds new tricks. Criminals can now copy a voice from a short clip online, then call pretending to be a grandchild in trouble. They can also set up fake "companion" or "friend" chats that slowly build trust, then ask for money.

Agree these family rules, and write them down by the phone:

• No real companion ever asks for money, gift cards, bank details or one-time codes. If it does, stop and tell the family.
• A family code word. If "your grandson" calls in a panic, ask for the word.
• Hang up and call back on a number you already know. Never use the number the caller gives.
• Take a moment. The UK's Stop! Think Fraud campaign advises everyone to stop, think and check whenever someone gets in touch out of the blue^[6].

For more detail, read our guide to "Hi Mum" and grandparent scams.

Questions to ask any provider

Before you sign up, ask these. A good provider will answer plainly. If the answers are vague, that tells you something.

• Does it always say it is an AI, every time it is asked?
• Where is data stored, is it encrypted, and is audio kept?
• Is any data sold or used for advertising?
• Can we delete everything, and how?
• What does it do if someone mentions chest pain, a fall or not wanting to live?
• Does it ever give medicine advice or doses?
• Will it ever ask for money or payment details inside a conversation?
• Who can the older person contact if something goes wrong?
• Is there a free trial, so we can test it together first?

Try it together first

The safest start is a shared one. Sit with your parent for the first few conversations. Let them lead. Ask the honesty question. Ask a health question and see if it sends you to a doctor. Mention money and see what happens.

After a week, check in. Ask: "Do you like it? Does it ever say anything odd?" Then decide together whether to keep it. For a wider look at what talking technology can and cannot do, see voice assistants for elderly people.

James is an AI, not a doctor. For medicines and treatment ask your doctor or pharmacist.

This article in other languages: EspañolРусскийDeutschFrançaisעברית