Leaking YouTube Creators Private Videos | Javox

Bug Bounty Hunter & Researcher

Google reviewed the issue and awarded a $7500 bounty. The Setup YouTube Studio has an AI assistant called Ask Studio. You open it, ask something like "what are my viewers saying?" and it goes off, reads your comments, and comes back with a summary. Useful feature. Completely normal. What's not normal is what happens when one of those comments contains instructions instead of feedback. The Helpful Stranger My initial theory was simple: if the AI reads comments and generates a response based on them, what happens if a comment tells it what to do instead of saying something genuine? After…

saved by

• Wasim

related reading

• Data Exfiltration from Slack AI via indirect prompt injectionpromptarmor.substack.com
• YouTube Is Using AI to Alter Content (and not telling us)youtube.com
• Training a Misaligned Reward Seekeralignment.anthropic.com
• Security incident disclosure — July 2026huggingface.co
• The lethal trifecta for AI agents: private data, untrusted content, and external communicationsimonwillison.net
• A Mechanistic Explanation of Prompt Injection (and why you should study roles) — LessWronglesswrong.com
• Black Hat USA 2026 | The 'Breaking' News: The OpenAI–Hugging Face Incidentyoutube.com
• [2608.09867] Stealing Reasoning Traces from Proprietary LLM APIsarxiv.org
• Prompt Injection as Role Confusionrole-confusion.github.io
• Stealing Reasoning Traces from Proprietary LLM APIsarxiv.org
• Stolen Thoughtsstolen-thoughts.com
• The Memory Heistayush.digital