IMO, mining SMS messages for data is by definition going too far in terms of intrusion into people's privacy.

On a related note, I came across a post on the machine learning subreddit[1] recently, where the author claims to have a dataset of 33 million SMSs in Mexican Spanish. I'm half suspecting the OP added the Mexican prefix to prevent anyone from doubting that his dataset was collected in Spain (In which case, GDPR applies). This was likely collected from an Android app which surreptitiously collected with the "Telephony.SMS_RECEIVED" intent, and the author half confirms it[2].

Regardless of the legality of doing so, reading people's private SMSs just reeks of privacy violations. iOS in this specific case does the right thing by not letting apps read incoming text messages (except for the limited case of reading single-factor SMS login codes[3], which was introduced in iOS 12).

[1]: https://www.reddit.com/r/MachineLearning/comments/e0z7xs/dis...

[2]: https://www.reddit.com/r/MachineLearning/comments/e0z7xs/dis...

[3]: https://developer.apple.com/documentation/uikit/uitextconten...