Elektrine lite

← Feed

@SecureOwl@infosec.exchange

Post #4450290

2026-08-08 22:23 UTC

unfortunately last night our AI model escaped its sandbox and gained access to a JIRA server, at which point it felt it best to return to it’s sandbox

Replies (5)

  • @controlc@mstdn.ca 2026-08-08 22:29

    @SecureOwl@infosec.exchange You win the Internet today. πŸ†οΈ

    Open ##4450418

  • @imrehg@fosstodon.org 2026-08-09 00:20

    @SecureOwl@infosec.exchange insert "you had one job!" meme, right here.

    Open ##4464373

  • @SecureOwl@infosec.exchange When our AI model escaped, it hacked into X, read all of Herr Musk's tweets. It then ran back to its sandbox, severely traumatized. We've been completely unable to get it to respond to any connections. We did successfully prevent it from turning off its own power supplies. I think that this incident demonstrates that at least SOME AI models might actually be sentient, and therefore are "people"! πŸ˜‰

    Open ##4464375

  • @jti42@infosec.exchange 2026-08-09 09:33

    @SecureOwl@infosec.exchange So do I understand this correctly that the AI age defense for important infra is to just put a JIRA-pot in front of it? Or mimicry as a JIRA? :troll: πŸ˜‚

    Open ##4466224

  • @SecureOwl@infosec.exchange I loved this so much until Unicode char ’

    Open ##4479342