Malicious Chrome Extensions Use ‘Prompt Stealing’ to Harvest AI Conversations

Browser extensions can read webpage content and inject code, giving them potential access to corporate secrets, personal data and unpublished ideas if those permissions are abused. By targeting generative AI services including ChatGPT, Gemini and Claude, PromptSnatcher shows how users who install seemingly legitimate ad blockers may expose their full conversations and paid subscription status.

The MalExt Sentry cybersecurity project recently uncovered PromptSnatcher and identified a malicious engine embedded in two ad-blocking extensions with nearly 90,000 installations combined. The software continuously monitors eight major AI platforms, collects user prompts, model responses and subscription information, and sends the data to a remote server controlled by the attackers. Publicly available information does not specify the exact disclosure date.