AI Use Statement

How we use generative AI tools to help write code and documents, and the safeguards that keep a named person accountable and client data protected.

1. Purpose and scope

This statement explains, in plain English, how Carefully Coded Ltd uses generative artificial intelligence (AI) tools, and the rules we follow when we do.

It covers:

  • AI tools we use in our own work
  • AI features in software we build for clients.

It applies to everyone who works for us, including contractors and subcontractors.

2. How we use AI

We use generative AI tools to help us work faster and more carefully. These include:

  • AI coding assistants, which suggest, explain and review code
  • AI writing assistants, which help us draft and edit documents.

We use them to help draft documents including bid responses, proposals, technical documentation and policy documents, including this one.

AI tools help us. They do not make decisions for us, and they do not replace our skill or judgement.

3. A named person is accountable for everything we produce

Every piece of AI-assisted work is reviewed, edited and approved by a named person, who is accountable for it as if they had written it from scratch.

  • For code, the approver is the engineer who merges it.
  • For bid responses, policies and other documents, the approver is the director, Tom Emmerson, unless another named person is recorded.

The approver must check that the work is accurate, complete and suitable. AI tools can produce confident answers that are wrong. We check facts, figures, legal references and claims against reliable sources before we use them. We do not let AI invent facts about our company, our experience or our certifications.

4. Protecting client and Authority data

Our clients include councils and other public authorities. We follow these rules for all client and Authority data:

  • We never enter client or Authority data into an AI tool that uses the data it receives to train its models.
  • We only use AI tools with model training on our data switched off, either through the provider's terms or the account's training opt-out setting. We check this setting and the terms before we start using a tool, and again when they change.
  • We never use client or Authority data to train, fine-tune or improve any AI model.
  • We do not enter personal data, security credentials, passwords or keys into AI tools.
  • Where a contract restricts or bans the use of AI tools, we follow the contract.

These rules work alongside our Data Protection Policy, Information Security Policy and Test Data Policy.

5. AI-assisted code

AI-assisted code is held to the same standard as any other code. Before it is released, it must pass:

  • human code review
  • our automated tests
  • our security checks, including dependency and vulnerability scanning
  • the same acceptance testing agreed with the client.

We check that AI-suggested code does not bring in unlicensed or unsuitable third-party code. The engineer who submits AI-assisted code must understand it well enough to explain and maintain it.

6. AI features in software we deliver

We will never put an AI feature into software we deliver to a client without the client's prior written approval.

An AI feature means any part of the software that uses an AI model, for example to generate text, summarise information, make recommendations or analyse images.

When we propose an AI feature, we will tell the client in writing:

  • what the feature does and which AI service it uses
  • what data it sends to the AI service, and where that data is processed
  • the main risks, including the risk of wrong answers, and how we will reduce them
  • how users will know they are using an AI feature
  • how a person can check or override what the AI produces.

7. Personal data and impact assessments

If an AI tool or feature will process personal data, we will:

  • carry out a data protection impact assessment (DPIA) before it is used, or help the client carry out its own DPIA where the client is the data controller
  • only go ahead when the risks have been reduced to an acceptable level and the client has agreed
  • make sure no one is subject to a decision made solely by AI that has a legal or similarly significant effect on them, unless the law allows it and the client has approved it.

8. Mistakes and incidents

If AI-assisted work causes a problem, such as an error in a delivered document or a security weakness in code, we treat it like any other issue. We fix it, tell the client where it affects them, and learn from it. If client data is entered into an AI tool against these rules, we treat it as a security incident under our Incident Response Procedure.

9. Questions

If you have a question about how we use AI, or want to know whether AI was used in a piece of work we produced for you, contact our director, Tom Emmerson, at tom@carefully.codes. We will give you a straight answer.

10. Review

AI tools and the law around them are changing quickly. The director reviews this statement at least every year, and whenever we start using a new type of AI tool.

See also our Accessibility Statement.

Organisation
Carefully Coded Ltd
Version
1.0
Owner
Tom Emmerson, Director
Effective from
19 November 2026
Next review
19 November 2027

Questions about this policy? Email info@carefully.codes.