Overview
- The prosecution case at Lewes Crown Court collapsed after the complainant admitted using an AI chatbot to prepare cross-examination answers, a development first reported on Monday that led to disclosure of her chat logs.
- Screenshots from the complainant’s phone showed AI-generated material organised under headings such as 'On what happened that night', 'On Consent and Capacity', 'On Physical Evidence' and 'On Your Behaviour After', which prompted defence challenges to the truthfulness of her testimony.
- The defence argued that the court could not tell how far the witness’s evidence had been shaped by the chatbot, and the Crown Court concluded a fair trial was no longer possible, halting the prosecution.
- The Court of Appeal held that using AI to prepare substantive answers can amount to prohibited 'witness coaching', urged that witnesses be firmly discouraged from using tools like ChatGPT, Claude or Gemini, and directed a retrial at the earliest opportunity.
- Judges said AI poses two risks—an honest witness’s memory can be unconsciously reshaped by suggested wording and a dishonest witness can use AI to refine false answers—so courts will need new rules and safeguards for generative AI in evidence preparation.