Cookies on the Defra digital service manual

We use some essential cookies to make this service work.

We'd also like to use analytics cookies so we can understand how you use the service and make improvements.

View cookies
Skip to main content
Department for Environment, Food and Rural Affairs: AI digital toolkit
Deliver with AI

Ethics

Using AI responsibly means staying accountable, preventing harm, and protecting people's data. These principles apply to every project.

Stay accountable

You are responsible for everything AI produces on your behalf. If the output is wrong, biased, or harmful, the responsibility sits with you and your team.

Do not treat AI as an authority. Treat it as a tool you must supervise.

Prevent bias

AI models reflect the data they were trained on, which can include biases.

Test your AI outputs with diverse scenarios and edge cases. For a Defra service that means testing Welsh-language and assisted-digital users, not just the mainstream path.

Look for patterns that disadvantage particular groups. Challenge assumptions in the training data and in your prompts.

Protect data

Follow UK GDPR and the Data Protection Act 2018.

Use only approved tools that meet Defra's security standards. Do not enter sensitive, personal or classified data into public AI tools.

The mechanics, including when a Data Protection Impact Assessment is needed, live in Keeping data safe.

Ask these 5 questions

Before using AI on a task, ask yourself these 5 questions:

  1. Does this serve a genuine user need? AI should solve real problems, not be used for its own sake.
  2. Can I explain how this works? You should be able to describe what the AI does and why you trust its output.
  3. Am I prepared to accept responsibility? If something goes wrong, you must own the consequences.
  4. Have I considered bias? Think about who could be affected and whether the output could be unfair.
  5. Does this protect people's data? Confirm that no sensitive information is exposed or mishandled.

If you cannot answer "yes" to all 5, change your approach.

Public trust and transparency

Government services depend on public trust. AI introduces new ways for that trust to be lost, through unexplainable decisions, biased outputs or mishandled data. Be transparent about where AI is used in services that affect citizens.

If your tool significantly influences a decision with a public effect, or interacts directly with the public, you must publish a record in the Algorithmic Transparency Recording Standard (ATRS). This is mandatory across central government.

Talk to the AI Capability and Enablement team (AICE) if you are not sure whether your tool is in scope.

Get help with this

Ask the AI Capability and Enablement team for advice or hands-on support.