The Digital Press

New AI-powered web browsers like OpenAI’s ChatGPT Atlas and Perplexity’s Comet offer agents that perform tasks by interacting with websites, but they introduce significant privacy and security risks.

Why it matters: AI browser agents require deep access to user data, increasing privacy exposure and vulnerability to malicious attacks.

The stakes: Prompt injection attacks can trick agents into leaking sensitive info or performing harmful actions without user consent.

The big picture: This security challenge is systemic and unresolved, prompting companies to implement partial safeguards amidst ongoing cat-and-mouse dynamics.

Commenters say: Users appreciate the convenience but express concern about privacy risks, urging caution and advocating for stronger protections before broad adoption.