I do see the copyright/intellectual property angle of training LLMs on the entire web, but what's the privacy issue here?

If you publish something on the web, what are you expecting to happen?

You should try us :) open-source and privacy-first alternative to Atlas -- https://github.com/browseros-ai/BrowserOS

Seems like it's based on Chromium? If so, that's a no-go for me. We need more web diversity and support smaller browser engines, we don't need yet another Chromium/Blink based browser.

I agree with this sentiment, but besides Webkit/Chromium and Firefox's Gecko, there's not really any options. Ladybird is a new implementation gaining fast though don't think it's ready to replace everyday workflows yet. And Ladybird has been a huge undertaking of course.

Building a new browser engine is 99% of the work and slapping LLM features on it is the other 1% of it.

Copying public data isn’t a breach of anyone’s privacy.

If they're the browser then they can see non-public data since people will be browsing those pages.

Chrome for sure