Researcher Demonstrates Prompt Injection Vulnerability in YouTube Studio AI (javoriuski.com)

• YouTube Studio AI can be manipulated via comments due to lack of separation between user content and system prompts.
• An attacker can leave a comment that tricks the AI into outputting attacker-controlled text, and even construct links that exfiltrate private video titles.
• Google classified the report as social engineering and declined to fix; the researcher believes it is a trust model violation.

"A security researcher discovered a prompt injection vulnerability in YouTube Studio's Ask Studio AI assistant. By leaving specially crafted comments on videos, an attacker can make the AI output attacker-controlled text, including links that leak private video titles. Google dismissed the report as social engineering, but the researcher argues it exploits trust in Google's product. The vulnerability remains unpatched."