An artificial intelligence assistant is a genuine convenience. It helps you write a difficult letter, understand a document, organise a busy week. Nobody wants to give that up for everyday tasks.
The problem starts when the same convenience spreads without you deciding: the history keeps years of conversations, the memory gathers details about you and the people you mention, a shared link stays readable, a connected app keeps reaching your email. Usually nothing serious happens. Sometimes something happens that takes time, money or patience to put right.
This post describes what sharing with an AI assistant can lead to: what can happen, in practice, when an assistant’s settings stay as they were on day one. It is the practical side of the recommendation to check your AI assistant’s data settings. The technical effects underneath, meaning what gets hit in terms of confidentiality, accuracy and availability, are explained in the post on AI chatbot privacy risks.
A realistic scenario
John is 68, retired and lives with his wife. For a few months he has been using an AI assistant on his tablet, connected to his email, to deal with household matters. In the spring there is a disagreement with the managing agent of his block of flats over the cost of the roof repairs, and John asks the assistant to help him write a clear, polite letter.
To explain himself properly, he pastes the managing agent’s email into the chat, uploads a photo of the residents’ meeting minutes with the names and signatures of his neighbours, and adds his address and phone number for the signature. The letter comes out well. John is pleased and, to show it to the other residents, uses the share button: the service creates a link to the conversation, which he posts in the neighbourhood group on a social network, open to everyone.
Two weeks later a neighbour messages him, worried: her name and signature can be read by anyone who opens that link. Around the same time John receives a text that seems to come from the managing agent, with his exact address, asking him to settle “the outstanding roof contribution” with an urgent bank transfer. John does not pay and phones the managing agent on the number he knows: the message is fake. The managing agent, who knew nothing about it, warns the other residents, and for a few days it is all anyone in the building talks about.
Nobody “attacked” his tablet. The conversation, with everything in it, became public through a link, and someone used the details they found there.
That evening his daughter helps him put things right. They remove the link in the settings and delete the post in the group, delete the chat, open the assistant’s memory and find the line “John has a dispute with the managing agent”: they delete it. They also look at the connected apps and remove the email permission to send messages, which John did not use. John decides on two simple rules: I do not upload other people’s documents in full, and I show a conversation by copying the text I need, not by sharing the link.
1. Operational consequences: when time goes on putting things right
A practical example
A self-employed tradesman uses an assistant with his personal account to prepare quotes and has connected it to his email, with permission to send. One afternoon he asks it to “draft replies to customers”, and a draft with one customer’s prices goes to another.
Possible effects
- apologies and explanations to the customers involved, and a review of every email sent;
- a check of which connected apps have which permissions, and removal of the excess ones;
- a review of the history and the memory, to understand what the assistant knows about customers;
- time lost rebuilding documents deleted in a hurry.
Why it matters
The damage is rarely serious, but the work of putting things right is real, and it often involves people who never chose to be part of your conversation with the assistant. And it almost always starts from a permission granted once, for convenience, and then forgotten.
2. Financial consequences: when details become bait
A practical example
As happened to John, the details of a conversation that has become public, an address, a dispute about the building, the managing agent’s name, allow someone to write a very believable fake message.
Possible effects
- scam attempts built on real facts from your life;
- costs and time to block a card or dispute a payment, if the attempt succeeds;
- a subscription started by mistake by a connected app with permission to make purchases.
Why it matters
The more a message knows about you, the more genuine it looks. A private conversation contains exactly the details that make a scam believable: keeping it private, and limiting what the assistant keeps, takes material away from anyone looking for it. Faced with an unexpected request for payment, the rule remains to check through a channel you already know.
3. Legal and regulatory consequences: when the data belongs to others too
A practical example
The minutes John uploaded included other people’s names and signatures. In other cases people paste a candidate’s CV, a customer list or a colleague’s emails into a chat.
Possible effects
- other people’s data kept by a service they did not choose;
- at work, a breach of the organisation’s rules on using AI tools;
- difficulty getting removed what has already been used: on many services, turning training off only covers future conversations.
Why it matters
Data protection rules give you practical tools. In the UK and the EU, the UK GDPR and the GDPR include the right to object (Article 21) when a service uses your data on the basis of its own legitimate interests, and the right to ask for erasure when the data is no longer needed or is being used unlawfully. The Information Commissioner’s Office (ICO), the UK’s data protection regulator, explains that organisations must respond to an objection within one calendar month, which can be extended by up to two months for complex requests, and that if you are not satisfied you can raise a concern with the ICO; in the EU, you can complain to your national data protection authority. This post is for general information: for a specific case, contact the data protection authority in your country or a professional adviser.
4. Reputational and relationship consequences: when the memory speaks for you
A practical example
An office worker shares her screen during a meeting to show colleagues how the assistant summarises a document. The answer begins “As you told me, you are preparing for the interview with the other company…”. The memory had kept a detail from a chat weeks earlier.
Possible effects
- personal information revealed to colleagues, relatives or acquaintances;
- embarrassment, or questions you would rather not have been asked;
- people mentioned in your chats, like John’s neighbours, who feel exposed.
Why it matters
Memory is meant to be useful to you, not to be seen by others. But it shows up in the answers, and answers are read aloud or on shared screens. Rereading it now and then is a simple way to decide what it can say.
5. Personal consequences: when you feel less free
A practical example
Someone notices that the assistant often suggests topics linked to a rough patch at work a few months earlier. They do not remember asking for that. They feel more “known” than they wanted to be.
Possible effects
- the feeling that the tool knows more about you than the people close to you;
- the temptation to stop using it, even for the things it genuinely helped with;
- after an episode like John’s, wariness of every “share” button.
Why it matters
Privacy is also the freedom to ask a question without it becoming part of a profile. The answer is not to give up the assistant, but to take back the choice: decide what it remembers, how much it keeps and whether your conversations may help improve the model. Leaving training on, for example, is a legitimate choice, as long as it is yours.
There is a subtler effect too: trust in the tool. An assistant that “knows” things you do not remember telling it gives the impression of not really being at your service. A memory you have reread, a history that lasts as long as you chose and few connected apps give the opposite feeling: you know what it keeps about you, and why.
The cost nobody budgets for: time
The most common cost of the consequences described here is not money. It is time.
| Situation | Typical time to sort it out |
|---|---|
| Checking the data settings, before anything happens | 15 minutes per assistant, once a year |
| Removing a shared link and deleting the post where you published it | 10–20 minutes, if you remember where you put it |
| Rereading and correcting the memory | 5–15 minutes |
| Dealing with a draft sent by mistake through a connected app | from an hour to a day, with explanations to the recipients |
| Recovering an assistant account used by someone else | from a few hours to several days |
The first row costs a few minutes. The others, a great deal more.
Consequences that fall on others
What you share with an assistant is not only about you.
- The people you mention. Other people’s names, emails and documents go into the history and, sometimes, the memory, without them knowing.
- Family members using the same device. On a household tablet, whoever opens the assistant sees your history and answers shaped by your memory.
- Colleagues and the organisation. A work document pasted into a personal assistant leaves the rules the organisation has set.
- Your email contacts. A connected app with permission to send can send messages in your name to anyone.
How this connects to the recommendation
All the consequences described here share the same starting point: settings accepted on day one and never opened again. Recommendation R38 works right there. Deciding how long to keep your history, rereading the memory, choosing on training and disconnecting the apps you do not need does not remove every risk, but it takes many of the choices that matter out of the hands of chance.
It is worth noting what you do not need. You do not need to understand how models work, nor to stop using the assistant. The difference between John’s scenario and any ordinary letter comes down to two actions: uploading only the useful part of the document, and showing the letter by copying its text rather than sharing the whole conversation.
How to reduce the risk
- Check the settings when you start using an assistant, then once a year or when the service changes its rules.
- Decide how long to keep your history and delete the chats you do not need.
- Reread the memory and delete what concerns other people or what you do not want kept.
- Choose on training, knowing that the choice works going forward.
- Use shared links with care: anyone with the link can read the conversation. Review the list of active links, if the service shows one.
- Connect only the apps you use, with the minimum permissions: read rather than send, when that is enough.
- Upload only the useful part of documents, especially if they contain other people’s data.
The full sequence, from prevention to recovery, is in the guide on how to delete AI chat history and memory. What is worth writing in the first place is in the resource on what not to share with AI chatbots.
How this connects to the Cyber Welfare Framework
| Pillar | What this content contributes |
|---|---|
| Awareness | Recognising that a conversation can affect time, money, rights and relationships, even when nobody “attacks” the device |
| Skills | Being able to trace a problem (a public link, a draft sent, a memory out of place) back to the setting that made it possible |
| Secure Behaviour | Checking the settings regularly, sharing the text rather than the link and connecting only the apps you need |
Reference level: FL2 — Beginner. Understanding the concrete consequences is what turns a forgotten setting into a deliberate choice. To see where you stand on your other digital habits, take the digital resilience self-assessment.
Quick checklist
- ☐ I know how long my assistant keeps its history
- ☐ I have reread the memory and deleted details about other people
- ☐ I have decided, deliberately, whether my chats may be used for training
- ☐ I have no active shared links I no longer need
- ☐ Connected apps have only the permissions I use
- ☐ I upload only the useful part of documents
Conclusion
Default settings are not the enemy. They are designed to make the assistant useful straight away, and over time they stretch to everything you give it. The consequences, when they come, are almost always manageable, but they cost time, sometimes money, and some peace of mind, yours and that of the people mentioned in your conversations.
Reopening the settings once a year is a simple way to keep the convenience where it helps and the choice where it belongs: with you.
Related content
- Check your AI assistant’s data settings — the recommendation this post builds on
- AI chatbot privacy risks — what gets hit, from confidentiality to availability
- How to delete AI chat history and memory — the steps to put things right and prevent problems
- Prompt injection attacks — how chats, memories and connected apps are targeted
- Consequences of a hacked social account — the related topic of the domino effect on connected services
Related resources
Short reads from the Resources section, for anyone who wants to focus on a single aspect:
- What Not to Share with AI Chatbots: Keep, Generalise, Share
- Opt Out of AI Training: Check Your Chatbot Settings
Start with the first step: the Cyber Welfare Programme guides you free of charge, one recommendation at a time.




