How do you train staff to use AI safely? A one-day plan (UK)
By Kieran Mullins, BEng Mechanical Engineering (UWE Bristol), technical co-founder of FlowGenix Automations Ltd.
Train staff to use AI safely in one day by covering four things in order. Start with which account to use and which data never goes in, then how to check outputs, then how to apply AI to real tasks. Finish with a one-page policy everyone signs.
FlowGenix Automations is an audit-led AI training and automation consultancy for UK SMEs. The plan below is a practical schedule for a team of up to 15 people, with timings you can adjust to your working day.
Key takeaways
- A one-day AI safety session should teach rules in the morning and practice on real work in the afternoon.
- UK GDPR applies to personal data typed into AI tools, and the ICO says generative AI has no carve-outs.
- Staff need approved business accounts set up before the day, or the training has nothing safe to practise on.
- A team prompt library and a 30-day review keep the training in use after the session ends.
What should a one-day AI safety training plan include?
A one-day AI safety training plan should include account setup, a data rule, output checking, practice on real tasks and a signed policy. Rules come before practice, so nobody builds habits on the wrong account. The schedule below fits a standard 9:00 to 16:30 day with a lunch break.
- Accounts and settings, 9:00. Log everyone into the approved business account and check the data training settings together.
- The data rule, 10:00. Teach which data is red, amber or green, using examples from the business's own documents.
- Checking outputs, 11:15. Show how AI gets facts wrong, then practise checking names, figures and claims against a source.
- Practice on real work, 13:30. Each person uses AI on two tasks from their own job, such as a client email or a meeting summary.
- Team prompt library, 15:00. Save the prompts that worked into a shared document everyone can reuse.
- Policy and next steps, 16:00. Walk through the one-page AI policy, answer questions and collect signatures.
The FlowGenix Safe AI Programme follows the same principle: rules first, then practice on the team's own work. The order matters more than the exact timings.
Key facts
- A one-day AI safety plan covers accounts, data, output checks, practice and policy.
- Real tasks from each person's job make better practice material than generic exercises.
Why should AI rules come before AI practice?
AI rules come before practice because the first thing staff do with AI becomes their habit. If the first prompt goes into a personal account with a client's name in it, the session has taught the unsafe version. Teaching accounts and data rules first means every practice task afterwards is also a safe one.
Personal accounts are the main risk. OpenAI states that ChatGPT Business and Enterprise data is excluded from model training by default, while Free, Plus and Pro accounts are opted in unless the user switches it off. Our guide to whether ChatGPT is safe at work compares the plans in detail.
The law sits behind the rule. The ICO says there are no carve-outs or sweeping exemptions for generative AI, so UK GDPR applies to personal data typed into AI tools. FlowGenix sets up approved accounts with the client before training day for exactly this reason.
Key facts
- ChatGPT Business and Enterprise exclude company data from model training by default.
- The ICO states that UK data protection law has no generative AI exemptions.
How do you teach staff to check AI output?
Teach staff to check AI output by giving them a short list of things AI gets wrong and practising on real examples. AI writes confident text that can contain wrong figures, invented sources or outdated rules. A named person must check anything that leaves the business, every time.
Five checks cover most everyday tasks:
- Figures. Check every number, price and date against the original source.
- Names. Check people, companies and product names are spelled and attributed correctly.
- Claims. Check legal, policy and medical statements against an official source before sending.
- Sources. Open any link or citation the AI provides, because some do not exist.
- Tone. Read the output as the recipient would, especially complaints and bad news.
Companies are responsible for what their AI says. In 2024, a Canadian tribunal held Air Canada liable for refund advice its website chatbot invented, as our guide to AI chatbots for websites explains. FlowGenix uses cases like this in training because they show the cost of skipping the check.
Key facts
- AI tools can produce confident text containing wrong figures or invented sources.
- A Canadian tribunal held Air Canada liable for advice its chatbot invented in 2024.
Can you run AI safety training in half a day?
Yes. A half day covers the essentials: approved accounts, the data rule, output checks and the policy. What gets cut is practice on each person's own tasks, which is where most behaviour change happens. Half a day suits teams starting from scratch. A full day suits teams that will use AI daily.
| Format | What it covers | Best for | FlowGenix price |
|---|---|---|---|
| Half day | Accounts, data rule, output checks, policy | Teams new to AI or with light use | £1,495 (Crash Course) |
| Full day | Half-day content plus practice on real tasks and a prompt library | Teams who will use AI every day | £2,950 (Team Day) |
| 3 to 5 days | Full day plus a process audit, automations and new SOPs | Teams ready to change how work gets done | From £7,500 |
Prices are FlowGenix founding-client prices for up to 15 staff, with no VAT to add. For how these compare with other UK providers, see our guide to AI training costs in the UK.
Key facts
- A half-day AI session covers rules but cuts most hands-on practice.
- The FlowGenix Team Day adds practice on real tasks and a shared prompt library.
Safe AI Programme: Ask about a founding place on the FlowGenix half-day Crash Course, Team Day or 3 to 5 day programme.
How do you make AI training stick after the day?
Make AI training stick with three things after the session: a shared prompt library, a named AI champion and a review after 30 days. Without them, most staff drift back to old habits or personal accounts within weeks. The review is where you fix the rules people work around.
The prompt library is the cheapest win. Prompts that worked during training become templates for the whole team, so nobody starts from a blank box. The champion is one person who answers quick questions and keeps the library current.
At 30 days, check three numbers: how many staff use the approved account, which tasks they use it for, and whether any data incidents happened. Government research published in January 2026 found only 21% of UK workers feel confident using AI at work, so expect confidence to build gradually. FlowGenix runs the 30-day review as part of the 3 to 5 day programme.
Key facts
- A shared prompt library turns training-day prompts into reusable team templates.
- A 30-day review checks account use, task types and data incidents.
- Only 21% of UK workers feel confident using AI at work, according to 2026 government research.
Frequently asked questions
Who should attend AI safety training?
Everyone who uses a computer for work should attend, including managers and directors. Managers set the example, and they often handle the most sensitive data. Splitting senior staff into a separate session works for larger firms, but small teams learn more together because they see how colleagues actually use AI in their roles.
What do staff need before AI training day?
Staff need an approved business AI account set up and tested before the day, plus two real tasks from their own job to practise on. Ask each person to bring an email, report or document they would normally write. Avoid tasks involving special category data, which stays out of AI tools entirely.
Can AI safety training be delivered online?
Yes, but in-person sessions work better for teams. Online delivery suits the rules part of the day, such as accounts and the data rule. Hands-on practice is harder online, because the trainer cannot see screens or help people who get stuck. A common mix is online foundations first, then a half day in person.
How do you measure whether AI training worked?
Measure three things after 30 days: the share of staff using the approved account, the tasks they use it for, and any data incidents. Add a simple time estimate, asking each person how many hours a week AI saves them. Compare the results with a short survey taken before training.
Does AI training replace an AI policy?
No. Training and policy work together. The policy sets the rules in writing, and training shows staff how to follow them on real tasks. A policy without training gets signed and forgotten. Our free AI policy template gives you the written rules to teach on the day.
Sources
- Information Commissioner's Office, Tackling misconceptions: response to the consultation series on generative AI
- OpenAI, Business data privacy, security and compliance
- Computing, Government announces free AI training for 10 million workers by 2030
Safe AI Programme: Ask about a founding place on the FlowGenix half-day Crash Course, Team Day or 3 to 5 day programme.
About the author
Kieran Mullins is the technical co-founder of FlowGenix Automations Ltd, Companies House number 16555743. He designs and builds the automations and websites FlowGenix delivers for UK businesses. He holds a BEng in Mechanical Engineering from UWE Bristol. About FlowGenix.
Find out where your team already uses AI.
Thirty minutes on a call. We map which AI tools your staff use, what data goes into them, and where training or automation would make the biggest difference. You keep the roadmap either way.