OpenAI Rolls Out Web Agent, Raises Privacy Warnings

Crypto News AI

Written by Aaron S.,

Editor-In-Chief

Last Updated: July 25, 2025

Key Takeaways

OpenAI’s new ChatGPT agent can browse sites, read emails, and manage accounts for users;
The agent may be vulnerable to prompt injection, where hidden commands trigger unwanted actions;
OpenAI warned that misuse could expose emails, files, or account settings to outside access.

Stop overpaying - start transferring money with Ogvio. Sign up, invite friends & grab Rewards now! 🎁

OpenAI Rolls Out Web Agent, Raises Privacy Warnings

BITDEGREE IN YOUR SOCIAL FEED

OpenAI has launched a new ChatGPT feature for its Plus, Pro, and Team subscribers that allows the chatbot to carry out tasks on websites automatically.

This update enables the assistant to perform tasks such as signing in to accounts, reading emails, making bookings, uploading files, and even connecting with services like Gmail, Google Drive, and GitHub.

However, OpenAI stated in a July 25 blog post that using the agent might put personal data at risk, particularly due to "prompt injection". This type of attack happens when someone adds hidden instructions to content that the chatbot might read, like a blog post, website text, or an email.

What Are Oracles in Crypto? (Beginner Friendly Animation)

Did you know?

Want to get smarter & wealthier with crypto?

Subscribe - We publish new crypto explainer videos every week!

If the chatbot follows those instructions, it could take actions the user never intended, such as sharing private information or changing account settings.

In a blog post, OpenAI explained that once the agent is allowed to access connected websites, it may also be able to view sensitive information, such as emails, files, or login data.

The post also noted that attackers could try to trick the assistant by hiding commands in everyday text. If that works, the assistant might send data to an unauthorized user or perform actions without the user’s knowledge.

OpenAI first announced the agent on July 17. At the time, the full launch was planned for a few days later, but it ended up being released on July 24, along with an update to the ChatGPT app.

Recently, xAI, Elon Musk's artificial intelligence (AI) company, fixed the problems with Grok 4’s offensive responses. What did the company say? Read the full story.