I would have sworn that autosave was enabled by default in absolutely every software that has anything to save since like the 2000s

Possible that we’re thinking about different features? Like for Microsoft Word, if I save a file to disk, make an edit, then exit out without saving (hitting “cancel” when it asks if I want to save) the disk copy is left untouched. That’s how the most tools work as far as I’m aware. It does have crash recovery (which may or may not work better than LibreOffice’s crash recovery, no idea).

You’re weird. Autosave is the norm in 2024

I do support challenging the software design before blaming the user, but I feel like I’m being thrown through a bit of a loop here. Autosave, while not unusual, is still the minority behaviour - surely?

I’m checking through tools I have installed and can’t find much that autosaves - even Word (tested editing a local file) doesn’t seem to autosave as far as I can tell. And, to be fair to the software, I often don’t want to overwrite the disk copy automatically (though there are some “best of both worlds” approaches, like with VSCode).

It doesn’t have “memory” of what it has generated previously, other than the current conversation. The answer you get from it won’t be much better than random guessing.

To file an infringement suit they’d need to have paid registration for each work which, even for the exorbitantly rich, wouldn’t be remotely feasible for all logical arrangements of words/images. There’s probably not even enough space in the Universe or time until its heat death to generate and store all such images.

Even if they did, copyright doesn’t protect against against independently created works that happen to be similar or even identical - so they wouldn’t be exhausting some limited set of possible works by doing so.

they are looking for specific ones that are in a database

They could be looking for any images without your knowing - there’s no guarantee that those images came from a CSAM database.

Its not like they can create a hash for a guy letting his dog on a horse

They could trivially create a hash for a picture of a guy letting his dog on a horse (which would also include other very similar images).

I didn’t necessarily mean to claim that they can scan for a concept lacking a fixed image, if that’s what you’re saying. That would theoretically be possible with enough hashes, but impractical.

And how was it privacy compromising?

1. Anything could be added to the hashes with the user having no way to know what’s being searched for beyond “trust us”. This could be partially alleviated if, for example, the hash had to be signed by organizations in a combination of states that’d make it difficult to push through hashes for anything other actual CSAM (so not just Five Eyes)

2. Adversarial examples to intentionally set off the filter were demonstrated to be possible. Apple made it clear that there are types of content they’d be legally obligated to report once they became aware of, and it’d be well within a government agency’s capabilities to honeypot, say initially, terrorist recruitment material

3. Coincidental false positives are also entirely possible (ImageNet had some naturally occuring clashes) and can result in their employees seeing your sensitive photographs

4. The user’s device acting against the user cements other user-hostile and privacy-hostile behavior. “People could circumvent the CSAM scan” would be given as another reason against right to repair and ability to see/modify the software your own device is running

5. Tech companies erode privacy by flip-flopping between “sure we’re giving ourselves abusable power, but we’ll stand up to governments pressuring us to expand this” and then “well what were we supposed to do, leave the market?” when they inevitably concede