In October, OpenAI unveiled an AI browser dubbed Atlas, purportedly designed to help you “understand your world” and “achieve your goals” by putting ChatGPT front and center.
A special “agent mode” was meant to book flights or buy groceries online on your behalf without you ever having to lift your finger.
“The browser just changed,” OpenAI’s head of ChatGPT boasted at the time, calling it a major step in ChatGPT evolving into an entire AI operating system.
But the cracks started to show almost immediately. There were glaring cybersecurity concerns, including a major vulnerability to prompt injection attacks. Its agent worked at a frustratingly glacial pace — it took the browser ten full minutes to add three items to an Amazon shopping cart, as The Verge found out the hard way. It even ignored huge portions of the internet like the plague, highlighting ongoing legal battles over copyright.
