Journal

A no-training promise does not establish zero retention or no external processing: follow clinical information through inference, logs, support and deletion.

The July 2026 MHRA clarification separates documentation from medical-device functions, but organisations still need to control clinical deployment risks.

AI that gathers a history needs different evidence from AI that records one, including tests of questions, omissions, escalation and clinical handover.

Improved attitudes after AMIE are evidence about patient experience, not proof of appropriate reliance, clinical benefit or confidence in data handling.

Medical learners should practise gathering information independently, challenging AI summaries and explaining decisions, without treating AMIE as proof of deskilling.

AgentClinic shows why interactive medical reasoning needs different tests from static questions, with practical implications for AI evaluation and clinical learning.

A feature-level guide to the Amazon Connect Health London preview, its implementation requirements and the questions UK clinical organisations should ask.

Understand the separate privacy questions around external AI providers, platform storage, clinical records and internal improvement in TORTUS Health.

Compare TORTUS Health and Heidi II through the patient relationship, clinical work, business model, regional access and responsibility for follow-through.

Understand selective autonomy, the difference between accuracy and coverage, and why a safe-looking hand-back policy still needs a workable clinical service.

Routine healthcare requests can hide relevant clinical context, making transaction completion an incomplete test of agent performance.

Learn what PatientAgentBench tests in simulated patient conversations, how its automated scoring works and which clinical questions remain unanswered.