There has recently been a wave of AI-generated accounts signing up pretending to be real people.

[clickhere@mastodon.ie]

@FediTips Thank you! ️

[hikergeek@mas.to]

@JdeBP @davecb @FediTips

Unfortunately I read that was part of the problem. Existing bot accounts were inviting a lot more bot accounts. At least currently when someone is invited it bypasses some of the normal checks on new users. I also read that invites have been turned off on a lot of servers to help stop this.

Double check what I am saying here though.

[klu9@ohai.social]

@CkrnkFrnchMn @FediTips

Inhumanly high output, e.g. paragraph-long replies in 5 different threads in under 3 minutes

[stux@mstdn.social]

@FediTips Pff.. it's a lot

Tips for mods/admins:

- Look for weird email TLDs
- Displayname mostly contains: "Firstname Lastname Emoji"
- Look for invited and go upchain to ban them all
- Check IP and ban where needed
- Good luck

[occurrent@app.wafrn.net]

@2something@transfem.social @FediTips@social.growyourown.services

How are we all spotting ai/bot/inauthentic accounts? I'm not intuitively great at this and would love to hear peoples' takes on what the indicators are.

#i-do-not-look-at-slop-and-as-a-result-i-do-not-always-realize-what-slop-looks-like #ditto-for-other-types-of-nonsense

[lritter@mastodon.gamedev.place]

@FediTips i always do this. one of the few pleasures i have in life, telling on bots.

[2something@transfem.social]

@occurrent@app.wafrn.net @FediTips@social.growyourown.services

I'm not either.

On fedi, since I'm not a mod, I tend to follow the approach of reporting when I am suspicious or unsure, and letting the mods sort it out. This sometimes leads to @puppygirlhornypost2 saying things like

i have beth report duty ​​

But she's also said

ok i lied i have like 4 more beth reports but i digress. god i think beth should actually just maintain a page highlighting some of this lol. the shit she sends, gems.

On Matrix, I am a mod for two rooms, but recognizing AI bots hasn't been much of an issue. The spammers have been really really obvious about being spammers. The other users show they are legitimate users over time. The harder part is sussing out bigotry.

EDIT:

Oh yeah and I guess I've also modded wikis, but thus far I haven't had to make a decision about whether a user is an LLM. On @wiki@gts.rainverse.wiki we've only ever banned on user, and it wasn't for LLM use. On Pandora's Tale Wiki there were 17 spam-only accounts made in quick succession, and it doesn't matter whether they were using LLM or some other method of generating spam.

[occurrent@app.wafrn.net]

@wiki@gts.rainverse.wiki @puppygirlhornypost2@transfem.social @2something@transfem.social @FediTips@social.growyourown.services

This is a good approach. For me I haven't felt confident reporting much since i just can't think of anything that would make me reasonably suspicious? I think my brain just has a blank spot there. Like if an account looks like it's selling something with a million repeated posts that's obviously spam, but I struggle to recognize anything else.. It's good to know you are On The Case lol.

[everydaymoggie@sfba.social]

If you have your instance set to approve signups manually, you can check a few things:
- The join reason is copied from someone else's existing profile, or it sounds nonsensical because it's written by a script that just pulls random fill-ins from somewhere
- The physical location of the IP address either doesn't match their claimed location, or it's a Tor exit node, or a data center (some legit signups come from data centers, too, so this is just one thing to look at)
- Disposable email in combination with one of the above

@CkrnkFrnchMn @FediTips

[missgayle@urbanists.social]

@FediTips

You know, requiring a mere $1 a month ($5 is better) per account of support for the instance in order to have an account (payable in advance - no recurring pymnt, no acct approval, and suspension if the recurring pymnt is stopped) solves both the bot problem and the freeloaders problem. Just saying. There's no tech solution that will weed out dishonest registrations faster or more efficiently.

[missgayle@urbanists.social]

@RaymondPierreL3 @FediTips

The "so what" is that as bandwidth and stirage costs continue to rise instances can't survive without funding, and it is morally and ethically correct to say that everyone who can contribute, should.

Nobody owes anyone a free platform without obligations. How is that different from predatory Epstein class behavior, take take take - but accept no responsibility for the or to the community?

[everydaymoggie@sfba.social]

We disabled them on our instance because they were bypassing the normal approval process for signups.

@HikerGeek

@JdeBP @davecb @FediTips

[ingonymous@mastodon.social]

@stux @FediTips
If you fight automated signups with manual tasks you will loose the game, no matter what

A small signup fee maybe could solve two problems at the same time?

[benjamin@piaille.fr]

@ingonymous @stux @FediTips I thought that before but now I disagree:
You can manually forbid subscription from some ISP entire networks (we banned all of Vietnam and Nigeria on Piaille)
Or more subtle, require validation for some email domains (we did that for outlook.com emails for a few weeks, very efficient)
So Mastodon have quite good tooling for manual but massive moderation

[donamasta@mastodon.social]

@FediTips what should we look for?

[feditips@social.growyourown.services]

p.s. Mastodon admins may want to switch off invitations entirely as the AI signups seem to be exploiting the invitation system. The setting for user invitations is a bit hidden but you can find it with this guide:

https://fedi.tips/inviting-people-onto-mastodon-and-the-fediverse/#AdminInvites