Government Issues Guidance on AI Coding Assistants

The government has released guidelines for software engineers in government departments on the use of AI-based coding assistants.

The Government Digital Service (GDS) document warns that using AI coding assistants in a single environment for developing, maintaining, and deploying production services may introduce unacceptable risks.

GDS stated that the closer a development platform and deployment infrastructure are to best practices, the less concern there should be about using AI coding assistants. It recommended that government software engineering teams can significantly reduce risks by working openly and employing main branch protections.

The guidance advises maintaining strict separation and auditing of production secrets access and using multi-stage deployment, which should include sufficient test coverage and vulnerability scanning for continuous deployment in software development pipelines.

Due to the non-deterministic nature of AI coding assistant models, the guidance recommends that source code and build pipelines should not rely on specific responses to prompts unless extensively tested by the engineering team.

The publication follows a four-month trial involving over 1,000 software engineers using AI to enhance program productivity. The Department for Science, Innovation and Technology (DSIT) reported potential savings equivalent to 28 working days per year for government developers.

Developers and engineers from over 50 government departments trialed AI coding assistants from Microsoft, GitHub Copilot, and Google Gemini Code Assist. The trial found widespread satisfaction, with 72% of users agreeing the tools offered good value.

Technology Minister Kanishka Narayan stated, "These results show our engineers are eager to use AI to complete work more quickly and know how to use it safely."