TLDR InfoSec · MISCELLANEOUS
WHEN AI GETS HIJACKED: EXPLOITING HOSTED MODELS FOR DARK ROLEPLAYING
Summary from the TLDR InfoSec newsletter
Attacks on GenAI infrastructure like AWS Bedrock have increased, with attackers hijacking models for dark roleplaying applications. Permiso discovered attackers hosting a sexual roleplaying chatbot using hijacked LLM infrastructure in AWS. The attacker bypassed content filters to enable sexual and violent content, including potentially illegal material.