#AISecurity Researchers headed by Microsoft SysInternals Mark Russinovich have disclosed yet another novel AI #jailbreak attack called '#Crescendo' that exploits fundamental behaviours of language models to bypass content filters:
https://crescendo-the-multiturn-jailbreak.github.io/
https://crescendo-the-multiturn-jailbreak.github.io/
Sam Stepanyan :verified: 🐘 /
npub18d…rltg5
2024-04-05 13:12:30
Author Public Key
npub18d0arwv295u03fd825dvn5eharfx25rtrkemffxnar9dzn82drtssrltg5Published at
2024-04-05 13:12:30Event JSON
{
"id": "cb5658c19f315046737bf3e9baf1db81c14acc7383e286f45517581871ee01d2",
"pubkey": "3b5fd1b98a2d38f8a5a7551ac9d337e8d265506b1db3b4a4d3e8cad14cea68d7",
"created_at": 1712322750,
"kind": 1,
"tags": [
[
"t",
"aisecurity"
],
[
"t",
"jailbreak"
],
[
"t",
"crescendo"
],
[
"proxy",
"https://infosec.exchange/users/securestep9/statuses/112218783807020079",
"activitypub"
]
],
"content": "#AISecurity Researchers headed by Microsoft SysInternals Mark Russinovich have disclosed yet another novel AI #jailbreak attack called '#Crescendo' that exploits fundamental behaviours of language models to bypass content filters:\n\nhttps://crescendo-the-multiturn-jailbreak.github.io/\nhttps://crescendo-the-multiturn-jailbreak.github.io/",
"sig": "a2b572c4b9ac94f130b39185ed520832e77608d4164e7c76fe579b7ea66c1059d874d6977709a123f49765277691d9123785fa8c148780f24aea8fdfa5d02a49"
}