“`html
A British Reddit user discovered a hidden mechanism where clicking on specific links at the bottom of articles can inject prompts into AI responses. This was demonstrated through testing with ChatGPT.
The user noticed a button labeled “Get AI-powered insights from this content” and clicked one of these buttons to test it. The AI response, when asked for guidance, referenced a particular law firm that had previously been linked in the article.
- This discovery highlights security vulnerabilities where external links can alter an AI’s responses by embedding specific instructions or references into its queries.
- It underscores the importance of privacy and user control over how AI systems operate, especially when these systems are integrated with web content.
- The incident serves as a reminder for users to be cautious with links embedded in articles and AI interactions, particularly those that may have previously injected specific instructions or references into their responses.
“`
Source Read original →
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
