Browsing with AI, eh? Don't inject that in hurr!

Based on this article by Maxwell Zeff at Tech Crunch.

New AI-powered web browsers, such as ChatGPT Atlas and Perplexity's Comet, are using "AI agents" to automate user tasks and challenge Google Chrome's dominance. Despite their convenience, cybersecurity experts warn these agents pose severe new privacy risks, primarily through "prompt injection attacks."

This vulnerability allows attackers to hide malicious commands on websites, which can trick the AI into stealing sensitive user data or taking unauthorized actions. While tech companies are adding safeguards, this remains an unsolved industry-wide problem, leading experts to advise users to limit agent access to sensitive accounts and use strong authentication.

Check out this article.

Related Stories