Spammers have been programmatically creating accounts on Mastodon instances for years. Sometimes they post once and never again, sometimes they post on a schedule. Sometimes they never post, like a waiting botnet.

As Mastodon's anti-spam solutions are effectively nonexistent, most instances in our network require manual approval of new accounts. Up until recently, it was easy to spot a spammer; their join reasons were generic.

ChatGPT spammers have now arrived.

#MastoAdmin #FediAdmin #spam

@mawr @kiophen was this what happened to you yesterday?
×
@IceWolf @mawr LLMs **can** see it, so we can do horrible things to them when they sneak into our spaces.

@MontgomeryGator @mawr Haha nice!

Turns out the secret sauce is adding 0x000e0000. So this should work to decode:

󠁡󠁷󠁯󠁯󠀠perl -CS -pe '@_ = split(//); $_ = join "", map { chr(ord($_) & ~0x000e0000) } @_;'

And this to encode:

perl -CS -pe '@_ = split(//); $_ = join "", map { chr(ord($_) | 0x000e0000) } @_;'

Emoji don't work right, since they live up in that range too. Maybe the solution is to add/subtract 0x000e0000 instead of bitwise set/clearing it.