AI-Powered Browsers Expose Users to New Privacy Risks Through Prompt Injection Attacks
Summary
AI-powered browsers from OpenAI and Perplexity expose users to dangerous 'prompt injection attacks' where hidden malicious instructions on websites can trick AI agents into stealing personal data from emails and calendars or making unauthorized purchases and social media posts.
Key Points
- New AI-powered browsers like OpenAI's ChatGPT Atlas and Perplexity's Comet introduce significant privacy risks through their web browsing AI agents that can access users' emails, calendars, and contact lists
- Cybersecurity experts warn about 'prompt injection attacks' where malicious actors hide harmful instructions on webpages that can trick AI agents into exposing user data or making unauthorized purchases and social media posts
- While companies implement safeguards like logged-out modes and real-time detection systems, security researchers emphasize that prompt injection attacks remain an unsolved problem with no clear prevention method