oxhak on Nostr: Researchers unveil 'Deceptive Delight,' a technique to jailbreak AI models by ...
Researchers unveil 'Deceptive Delight,' a technique to jailbreak AI models by slipping in covert instructions during chats. This raises serious concerns about LLM security. #AI #Cybersecurity #AdversarialAttacks
Published at
2024-10-23 10:30:22Event JSON
{
"id": "c1d786ce41d337ccb96dc6eac3e35fb70a51e6640c4687c34f146081ceb6a2b6",
"pubkey": "81b26cb98224311ea520a9042bf9c7cc78d2725d0a99f9797afd9a8a35970aaa",
"created_at": 1729679422,
"kind": 1,
"tags": [],
"content": "Researchers unveil 'Deceptive Delight,' a technique to jailbreak AI models by slipping in covert instructions during chats. This raises serious concerns about LLM security. #AI #Cybersecurity #AdversarialAttacks",
"sig": "c3826671d65138a0df5b805917745e4dccf8c3deea15af24d7d08f09c27fb19f58b53fc4880379cdd6bc644bf111b2f8a602bcc492191f3c08d006b740617a6f"
}
