Twitter/X has automatically opted you in to training Elon Musk's Grok AI on everything you've ever posted on the service. Here's where they have stashed the hidden setting, and how to turn it off.
Find the right setting and you can say now to X's Grok AI being trained on your work
Musk calls his AI tool Grok, and Twitter/X says it's AI with humor, "a rebellious streak and an outside perspective on humanity" that makes it "a unique and entertaining companion."
Sure. Okay.
Twitter/X also says it was not "pre-trained on X data (including public X posts)," and being very generous about it. But, that claim is at best outdated.
It may not have been pre-trained on posts, but it certainly is being trained on them now. Without an announcement and certainly without your opt-in permission, Grok is now using your posts to train it further.
Grok AI is a premium feature, a subscription option. But while that limits who can use it, there are no such limits on whose posts it can be trained on.
There are two ways to stop it doing this, though. The first is to deny it permission online.
Twitter/X has not yet and probably never will never show you where that setting is, you have to be told.
You also have to do it on a Mac or PC and in a browser. It cannot be changed on a mobile device as of July 2024, because of course it can't.
How to stop Grok AI training on your posts
- On a Mac, go to Grok settings
- Under Data Sharing, untick the permission box
- Optionally, click Delete conversation history
It's probably not worth deleting your conversation history, but maybe do it anyway. The wording of the setting appears to say that once denied, Grok can't use any of your posts, whether new or historical.
But good luck getting it un-trained from anything it's already picked up. Either this setting works and your conversation history won't be touched, or it already has been and there's nothing you can do about it.
And, it's not clear what happens if somebody who has left the checkbox on retweets something that you post. They still don't have a functional press contact, so if anybody from there is reading our emails in the future and sees this, maybe shoot us an email about it?
There is the second option, which is to leave the service entirely. That at least seems a lot easier than it used to.
Ultimately, leaving may turn out to be the sole workable method of preventing Grok using your work, or at least preventing it to use any more. For while we have no option but to assume the deny setting will be respected, this is a hidden setting in an unannounced feature, and despite vague government moves, there is no policing of this issue.