Here’s why I won’t enable automatic deletion of Siri chats

Every time the new Siri Finally When it launches, it will be the most privacy-friendly AI chatbot, with an additional privacy feature reported yesterday.

We already knew that Apple’s deal with Google would mean that Siri would not be allowed to be used to train the Gemini model, and Bloomberg now reports an option to automatically delete our Siri conversations…

Although the new Siri is powered by Google’s Gemini, it will not run on Google’s servers. Apple said it would instead run AI functions on-device where possible and the Gemini model on its own Private Cloud Compute servers if necessary. Outgoing CEO Tim Cook said at the time:

“We believe we can unlock many experiences and innovate in key ways through collaboration. We will continue to operate on-device and in private cloud computing, while maintaining our industry-leading privacy standards.”

Yesterday’s report states that the new Siri app will offer us the ability to automatically delete our chats afterwards.

Another privacy gain with the new Siri app will be the automatic deletion of conversation history. On the Messages app, you can set your conversation history to automatically delete after 30 days or a year, or leave it indefinitely. The same options will exist in the new Siri app.

This looks good, and I think it’s a positive thing that the option is available – but I won’t enable it myself. Indeed, AI chatbots learn from the history of all our discussions, and this context can be extremely useful in our future interactions.

Claude is my current AI of choice, and I have trained her both explicitly and implicitly to reflect my own preferences. I gave him a number of standing instructions – for example, be succinct, use bullet points whenever possible, avoid flowery language, refrain from sycophancy, and always include links to his sources of factual information. Also, when he answers a question in a style I like, I give him feedback on it and he has definitely learned from this practice.

There were other examples where Claude used the contextual knowledge he had about me to tailor his responses. These were not things I told him explicitly, but rather information he gleaned from previous questions I asked and web searches he did in response to my queries. There were quite a few times when Claude cleverly used this context to answer my questions.

For this reason, I will not automatically delete any of my Siri chats. There is always the possibility of manually Delete individual chats if you’re concerned that Siri will retain sensitive data from particular sessions. I’ve also done this for chat sessions used when I’m helping a friend with something, and I don’t want it to get confused by thinking the context is about me.

By the way, one thing I learned very quickly with AI chatbots is that you have to start a new conversation for each topic. If you don’t do this, it processes all the contents of the current chat every time you add something to it, which uses a lot more tokens. It’s also much easier to refer to previous sessions and resume them as needed if each is a separate discussion in your sidebar or history. Claude generally does a good job of giving each thread a sensible name.

This will then be my approach – and you? Please share your thoughts in the comments.