How to Use AI Browsers Without Getting Hacked
For the past few days, I’ve been poking around every AI browser I could get my hands on. So far, I’ve performed general research tasks on Perplexity Comet and used ChatGPT Atlas to successfully navigate an Amazon checkout. I even spent some time familiarizing myself with the new Dia browser from the developers of Arc . As I've explored these browsers, I've been mindful of the many security risks to contend with: Prompt injection, where malicious AI prompts are hidden in a website or browser extension’s HTML source code, is the most obvious threat. But there are also cases of AI agents acting without a user’s permission to access your logged-in accounts. Moreover, AI browsers can leak data between browser tabs and hand over user credentials on clever prompting without even using any malicious code. But despite the risks, there are legitimate ways to experiment with AI browsers without compromising your privacy. In fact, most of these browsers have optio...