Dev.toDOM 내 숨겨진 텍스트를 통한 AI Agent 대상 Indirect Prompt Injection 위협 분석Prompt injection through website content: how AI agents can be manipulated by the pages they visitSecurityintermediate13 분 소요5일 전