GOV.UK chatbot gets smarter but slower as LLMs improve

The UK Government Digital Service (GDS) has run two public pilots of its GOV.UK Chat service, improving answer accuracy from 76% to 90% by leveraging newer LLMs and internal data science work. However, more powerful frontier models have increased response times to an average of 10.7 seconds, which users find too slow. GDS is exploring streaming partial answers to improve perceived speed while maintaining safety guardrails. The chatbot, built on Amazon Bedrock with Anthropic's Claude, successfully resisted 508 adversarial attempts and now supports clarification requests for ambiguous queries. Plans include integrating the chatbot into the GOV.UK app and eventually rolling it out across the full GOV.UK website.