This is pretty worrying, both that there appear to be significant medical issues with the chatbot advice, and that their product analytics aren't mature enough to be able to tell them that.

They claim the doctor who raised concerns ran ~2400 tests, with 20 legitimate errors. An error rate of "0.8%".

The doctor who raised the issues claims this is more like 800 tests run to completion (where you can even get results), and that many of these were repeated tests, so far fewer separate tests. This would suggest an error rate of more like 2.5-10%, depending on how many repetitions he was making.

Separate to this, Babylon published this data about his usage of the app in a press release without his permission. They say this is ok because it's not personal data, but while it likely doesn't count as health data I would say that it is personal data. That said I'm sure one could diagnose Munchausen Syndrome with it, so potentially it is health data. It is _very_ worrying that they have such a cavalier attitude to data protection.