The UK's Crown Prosecution Service (CPS) has developed an AI-powered Correspondence Drafting Tool, seeking to automate complex compliance processes while maintaining regulatory oversight and data security.

The CPS tool, built in partnership with NTT Data UK, utilises Microsoft OpenAI's GPT-4 models to automatically pre-populate correspondence templates with case information drawn from the organisation's existing case management system.

The CPS also looked at alternatives, such as programmatic solutions, however, due to the contextual nature of the correspondence, it was determined that Large Language Models are more suitable for correspondence drafting. Multiple LLM models were considered, however, OpenAI service was identified as the most appropriate meeting CPS security guidelines and standards.

"The tool does not make decisions in the prosecution process, it is used to aid correspondence drafting," the CPS states in its algorithmic transparency record, emphasising the critical distinction between AI assistance and automated decision-making.

The CPS said that such correspondence was previously drafted in Microsoft Word and involved manual transfers of information, which left potential for errors.

Every AI-generated draft undergoes multiple layers of human review and approval in accordance with CPS guidelines, ensuring accuracy and quality before any correspondence is sent.

The organisation identified hallucination and security as primary risks, implementing specific mitigations including comprehensive user training, strict access controls, and regular quality audits.

"Data is contained within the CPS environment. The data is stored within the tool until the correspondence is finalised, then the information is deleted," according to the transparency documentation.

Phased Implementation Strategy

Currently in beta phase with 30 users, the CPS is taking a measured approach to rollout. This phased strategy has processed 230 total requests with over 440,000 tokens, providing measurable data on system performance before wider deployment.

The organisation conducted extensive user research throughout the iterative development process and developed comprehensive training guides covering roles and responsibilities for the review and approval process.

The CPS worked with NTT Data UK Limited through a pre-established competitive framework, with the supplier providing expertise in AI model integration, natural language processing, and data management. The cross-functional delivery team included user researchers, designers, business analysts, and AI experts.