How are the new AI features in version 11 implemented in terms of privacy?
Answers do not need to be overly detailed, but should help us to decide whether, and which AI features can be turned on.
The general gist is the same as in eM Client 10 - the prompts are sent to OpenAI.
The information is sent only to the AI provider and is not stored or logged by eM Client anywhere. We will always do our best to protect your data and privacy as much as possible.
In eM Client 10 all AI features were on demand- ie. only if you click a specific button to use them (Generate text, Reply with AI).
In eM Client 11, this is still true for most features - Summarize, Generate About me, Analyze sent mails, Natural language search, AI Rules… all run only when you choose to.
If you want to “opt out” of specific ones, you can hide the buttons from the toolbars and keep only the ones you plan to use.
But a few new features are “on display”, applied automatically when you open a specific message - those are detect events/tasks from email and quick reply buttons (that offer 3 options on how to reply to a specific message based on context).
With this addition, we expanded the Settings section for AI, which now has the General section to turn on/off all AI features, then the “Features” subsection where you can turn off the on-display ones (and Summarize, since that one is not part of the customizable toolbar), including option to set this differently for each account.
We are planning to update the AI about page on our website soon to explain each feature in detail.
At first: Thanks for the quick, and detailed reply! Looking forward to the updated AI about page…
That sounds good and bad at the same time: Good, because it probably means eM Client is using OpenAI’s API, and bad because sending customer-related information (mail contents) as part of the prompt to “elsewhere” is no option for us as a small IT consultancy. If you are using the API, could you make the API root URI customizable? That would open up the possibility to use self-hosted LLM servers. I understand that you could hardly offer support for such configurations, as all kind of problems may arise using self-hosted AI servers. Still, it could satisfy strict privacy policies.
It unfortunately didn’t make it into eM Client 11.0 because the development ended up taking a bit longer than expected, but we do hope to add custom APIs for the AI features in one of the subversions of eM Client 11.
Correction to my previous reply - I checked in with the team, and while it’s not available in the BETA yet, we might be able to squeeze the custom 3rd-party API keys in before the official eM Client 11 after all. So stay tuned!