Harness, don't hide.
How to set up AI well, keep young people safe, and use it as a tool to serve rather than a master to rule.
- Word
- kabash, Genesis 1:28
- Means
- to bring under cultivation, to bring order
- Time
- About 20 minutes to read and set up
By Dan Gillespie. The views in this guide are solely those of the author. They do not represent Youth for Christ Northern Ireland or any other Youth for Christ organisation.
A gift to steward, not a threat to hide from
When people meet a new technology, they tend to do one of three things: ignore it, escape it, or engage with it. Head in the sand is the easy option. It is also the one that leaves young people to figure it out alone.
In Genesis 1, before the Fall, God gives humanity a calling: be fruitful, fill the earth and subdue it. The Hebrew word is kabash. It means to take what God has made and, under his authority, bring it into order so it becomes useful and fruitful. Work, creativity and innovation are part of our original calling, not a consequence of sin.
So the question is not whether we engage with AI. It is whether we steward it faithfully. If that sounds like an impossible task, good. It means you are taking it seriously.
The Reformers didn't invent it. They harnessed it to put Scripture in people's hands.
One voice could reach tens of thousands. Billy Graham's crowds heard the gospel through sound systems.
Another turning point. Same question: will it shape us, or will we steward it?
None of these changed the gospel. They changed its reach.
AI can produce information. Only God, through the power of the Spirit, can bring transformation.The line to remember
What is AI, actually?
Artificial intelligence is any computer system that does something we would normally call thinking: recognising, predicting, deciding or writing.
The term covers a lot. The chess computer of the 1990s, the face unlock on your phone and ChatGPT are all AI, but they work in very different ways. It helps to see them as circles inside circles.
- Artificial intelligence
- The whole field. Includes old systems where people wrote every rule by hand.
- Machine learning
- Instead of writing rules, show the computer thousands of examples and let it learn the patterns.
- Deep learning
- Machine learning using many-layered "neural networks", loosely inspired by the brain.
- Generative AI
- Deep learning that makes new things: text, images, music and video. ChatGPT and Claude live here.
How ChatGPT and Claude work
They are pattern machines. They have read an enormous amount of human writing, and they predict what words should come next. That is why they sound so human: they were built from us.
It picks a likely next word, adds it, and repeats, thousands of times a minute. Illustrative figures. Sounding human is not the same as being human: it doesn't know your young people, remember their birthday or pray for them.
From rules to agents: a short history
- Foundations
- 1950Can machines think?
Alan Turing proposes his "imitation game", now called the Turing test.
- 1956"Artificial intelligence" is named
A summer workshop at Dartmouth College coins the term and launches the field.
- 1958The perceptron
Frank Rosenblatt builds an early neural network that learns from examples.
- 1966ELIZA, the first chatbot
A simple script that mirrors your words back. People still confided in it, a warning that rings true today.
- Winters and rules
- 1970s–80sThe AI winters
Promises outrun results. Funding dries up twice. "Expert systems" built on hand-written rules have modest success.
- 1997Deep Blue beats Kasparov
IBM's chess computer defeats the world champion, mostly through brute-force search.
- Learning from data
- 2000sMachine learning goes everyday
Spam filters, recommendations and search ranking quietly learn from data the internet now provides.
- 2012The deep learning breakthrough
A neural network called AlexNet leaps ahead at recognising images. Powerful graphics chips make big networks practical.
- 2016AlphaGo
DeepMind's system beats a world champion at Go, a game thought to need intuition.
- 2017The transformer
Google researchers publish a new design for language models. It is the "T" in ChatGPT, and the engine of everything since.
- Generative AI
- 2018–20Large language models grow
Models trained on vast amounts of text start writing convincing paragraphs.
- 2022ChatGPT launches
Released in November, it becomes one of the fastest-adopted apps in history. Image generators arrive the same year.
- 2023The assistants arrive
GPT-4, Claude and Gemini. AI moves into phones, schools, offices and youth groups.
- Agentic AI
- 2024Models that reason and use tools
AI learns to think through problems step by step, search the web and operate a computer.
- 2025–nowAgents
AI that doesn't just answer but acts: planning tasks, using your apps and working on its own for hours.
- NextProactive agents and agent teams
Agents that live in the tools your team already uses, notice what needs doing, start work without being asked and hand tasks to other agents.
Most of us still use AI like a very clever search box: type a question, read the answer, copy what's useful. That's the generative stage. The technology has already moved on. The tools you use every day are quietly gaining the ability to act for you, and that changes what "using AI safely" means.
The three steps of the agentic stage
You ask, it answers. You stay in charge of every action.
"Write a permission form for Saturday's trip."You set a goal and it takes several steps to reach it: searching, using your apps, checking back with you before anything important.
"Plan the trip: find a venue, draft the form and put it in my calendar."It works in the background, notices what needs doing and starts on its own. Several agents may pass work between them.
"I saw three forms haven't come back. Shall I send a reminder?"At each step you hand over a little more. So the question shifts. In the generative stage it was "what am I typing in?" In the agentic stage it becomes "what can this reach, and what can it do without asking me?"
Chatbot or agent?
The newest shift matters most for how you set things up. A chatbot talks. An agent acts.
Chatbot
- Writes, explains, brainstorms
- You decide what to do with the answer
- Risk sits mostly in what you type in
Agent
- Reads your email, files and calendar
- Browses, fills in forms, sends messages
- Risk also sits in what it can reach and do
Give an agent the least access it needs. Don't connect it to inboxes, drives or apps that hold young people's information. Make sure it asks before it sends, posts, buys or deletes anything, and check what it did afterwards. If a web page or document it reads tells it to do something, that isn't you asking.
Five ground rules before you sign up
These come first because a setting can be changed later, but a young person's information cannot be taken back once it has been shared.
- 1You use it. Young people don't, through you.
This guide is for staff and volunteers. Don't set up accounts for young people or hand them your logged-in device.
- 2No young person's personal information. Ever.
No names, photos, contact details, school, health, family situation or anything that could identify them.
- 3Safeguarding stays with people.
Never type a disclosure or concern into an AI tool. Follow your safeguarding procedure and speak to your designated lead.
- 4Check everything it gives you.
AI states wrong things confidently, including about God. Test it against Scripture and trusted people (1 Thessalonians 5:21).
- 5Be open about using it.
If AI helped you prepare something, it is fine to say so. Modelling honest use teaches more than any rule.
Age limits, if young people ask
Young people will ask whether they can use these tools. Here is what each company's own rules say.
| Tool | Minimum age | Under 18 |
|---|---|---|
| ChatGPT | 13 | Needs a parent or guardian's consent |
| Claude | 18 | Not permitted |
If you ever use AI in a session with under-13s, OpenAI says an adult must be the one interacting with it.
Set up ChatGPT
On free and Plus accounts, ChatGPT can use your conversations to train its models unless you switch it off. It takes under a minute.
Turn off model training
This stops your chats being used to improve OpenAI's models. It applies to your whole account and your chats stay in your history.
Profile›Settings›Data controlsImprove the model for everyoneOFFDecide about Memory
Memory lets ChatGPT remember details about you across chats. Keep it off if you share a device or mix work and personal use.
Settings›Personalization›MemoryUse Temporary Chat for anything sensitive
Temporary chats don't appear in your history and aren't used for training. Start one from the top of a new chat.
Set custom instructions
Tell it once who you are and how to help, so you don't repeat yourself. A starter is below.
Settings›Personalization›Customize ChatGPTClear out old chats now and then
Settings›General›Delete all chats
Business, Team and Enterprise plans don't train on your data by default.
Set up Claude
Claude works the same way. On Free, Pro and Max plans, there is a single switch that controls training, and it also affects how long your chats are kept.
Turn off model training
With this off, your chats aren't used to train future models and are kept for a much shorter time. It only affects chats from now on.
Profile›Settings›PrivacyHelp improve ClaudeOFFTurn off location metadata
On the web, it's in the same Privacy page. On the phone app, location is controlled in your phone's own settings.
Location metadataOFFCheck your shared chats
Any chat you've shared by link is visible to whoever has the link. Remove the ones you no longer need.
Settings›Privacy›Shared chats›ManageUse Incognito for anything sensitive
Incognito chats are never used for training, whatever your other settings say.
Use Projects for repeat work
Put your session template, house style and ministry context in a Project, so every chat inside it starts with that background.
Paste this into ChatGPT's custom instructions or a Claude Project, then edit it to fit you.
I'm a Youth for Christ youth worker. I work with young people aged 11–18. Help me plan sessions, write admin and prepare resources. Use plain British English. Keep a Christian, gospel-centred perspective and quote Bible references so I can check them. If you're unsure of something, say so. Never ask me for details about individual young people.
What goes in, and what never does
Before you paste anything, ask one question: if this chat leaked, would anyone be harmed or embarrassed? Then use the traffic light.
Go
Use freely- Session plans, games and discussion questions
- Emails, reports and permission-form wording
- Translating public resources
- Poster, social and event ideas
- Exploring a Bible passage from different angles
Pause
Strip out details first- A pastoral situation, fully anonymised and generalised
- Funding bids with organisational figures
- Anything a colleague wrote
- Your own testimony or personal story
Stop
Never put in- Young people's names, photos or contact details
- Safeguarding concerns or disclosures
- Health, family or behaviour notes
- Passwords, bank or donor details
Formation needs friction
Using AI is like playing a video game with every cheat code switched on. You get to the end faster, but you skip the struggle that builds the skill. Some of that struggle is exactly what forms us.
What the research is showing
An MIT study had 54 adults write essays over four sessions while their brain activity was measured. One group used only their own thinking, one used a search engine and one used ChatGPT.
The researchers called this cognitive debt: you save effort today and pay for it later in learning and memory. But one finding gives real hope. When people built the skill first and brought AI in afterwards, connectivity and recall improved. Order matters. Think first, then use the tool.
If that is true for adults, consider a teenager whose prefrontal cortex is still developing. Elon University's Being Human in 2035 report asked experts what life will look like a decade from now, and loss of social intelligence was one of their concerns.
What you always hand off, you slowly lose the ability to do.
Ask before you prompt: is this something to hand off, or something I need to build?
Wrestling with words and walking with people builds pathways nothing can shortcut.
You don't wear a seatbelt because you are a bad driver. You wear it because you value your life. These settings work the same way: wisdom, not fear.
And one test for your own use: would you be happy sharing your chat history with a friend?
Talking with young people about it
Young people are longing for safety and belonging, acceptance and recognition, significance and purpose. A chatbot that is always awake and never judges can feel like it offers all three. If you only ever say "AI is bad", they will stop telling you how they use it. Be curious instead.
- "What do you use AI for?"
- "Have you ever asked it about God, or about yourself? What did it say?"
- "Did the answer feel true? How would you know?"
A young person's chat history can become the start of a conversation about faith. If someone tells you they talk to AI more than to people, don't shame them. Ask what it gives them, and look for ways the body of Christ can meet that need. If there are signs of serious loneliness or distress, follow your safeguarding and pastoral care procedures.
You are the light of the world. A town built on a hill cannot be hidden.Matthew 5:14
Your setup checklist
Work through this with your phone or laptop open. Your ticks are saved on this device only.
Let AI do the tasks, so we can give more time to the people.
Neither the printing press, the microphone nor artificial intelligence is the gospel. But each can become a way for the gospel to reach further. So let's harness it, not hide from it.
App menus change often. If a setting has moved, search the app's help centre for the toggle name.