Post

ok, deep breath… we're discussing *potentially* adding post editing to Bluesky this year.

i know. it's the #1 thing yall have been asking for. it's a Bluesky meme. every time we post something, people reply "edit posts wen?" we see you. we have always seen you. so why haven't we shipped it?

post editing is easy to build, but it adds a bunch of complexity and risk that's hidden until you look below the surface.

there are various safety and spam risks, but the core problem is this: someone writes "puppies are the best 🐶", racks up engagement, then edits it to say something horrible or swaps in a link to a malicious site. now people look like they endorsed it.

this gets especially ugly with reposts and quote posts. plus, post editing adds more complexity and required tooling for our T&S team. so if we do this, we'll need guardrails.

here's what we're thinking: 1) you have 5 minutes to edit a post 2) you can only edit each posts once 3) images and videos can’t be edited 4) a prominent "edited" badge appears on the post 5) version history is visible to everyone 6) your account has to be at least a week old

This is really, really not enough to guard against serious abuse issues. At the bare minimum, you also have to clear interactions or include versioning, and even that's not enough. I'm begging you, email me, I will redteam this for you, there are SO MANY invisible abuse cases

didn't he say that it would include versioning?

Ah, you're correct, my eyes slid over that because I just woke up -- my bad! Okay, correct that to "attach the interaction to the version", which is what *I* mean when I say versioning but want to make sure it's part of what they mean

oh, yeah, i agree with that.

bsky.app/profile/apar...

I mean the great thing about that is that bluesky has the technical ability to do it already. They can, at the very LEAST, provably attest as to whether a user stated that their intent was to respond to an old version of the post.

(note I have phrased this very carefully: a user can always doctor their post to make it look like they were interacting with a prior version of the content even after the edited version is up and all Bluesky can do is say "well, the edited version EXISTED at the time this post was seen by us")

oh, sure. but like if we ignore the analog hole...

Probably not a wise thing to do when you're doing product safety engineering.

i guess it just seems quaint when i can ask chatgpt to generate an image of me saying I love pancakes

Yeah, but everybody knows images can be easily doctored. The signals the site itself gives you are expected to be a lot more robust by users.

? i mean my thought process is that since they can handle edits by versioning the content id but not the record id, you couldn’t doctor something to look like you interacted with the wrong version outside of literally doing analog hole stuff (editing a screenshot, changing the dom, etc)

If the original CID is recoverable (eg, because another user liked/replied to the unedited post), you can forge a like/qp/repost to the original version even after it was edited.