OpenAI Dots Reads Your Email And Grabs Login Codes For You

OpenAI has started letting its Dots agents pull one-time login codes from a user’s connected email inbox and type them in, removing one of the last spots where AI helpers stall and hand the job back to a human.

Key Takeaways

• Dots can now fetch a sign-in code sent to a connected inbox and enter it on the website asking for it
• The feature only works if the user has connected an inbox and given permission
• Using a saved login for a new sign-in still needs the user to confirm
• Dots are OpenAI’s always-on agents, which keep working in their own cloud browser after the user logs off

The change started showing up in ChatGPT over the weekend, and screenshots of the new option were shared on Saturday. OpenAI has not yet listed it in its public release notes.

Why Login Codes Matter For AI Agents

Many websites send a six-digit code to your email before they let you in. A person copies it across in seconds, but for an AI agent it has been a dead end. The bot could browse, click and fill in forms, then stop cold at the sign-in screen and wait for a human.

With inbox access, a dot can find the latest message from that website, pick out the code and enter it before it expires. That makes it far more useful for chores like booking travel, ordering supplies or logging into work tools that use email checks.

Also Read: Axi Lets Retail Traders Bet On OpenAI And Anthropic Prices

What Dots Are

OpenAI introduced Dots on Sept. 29 as agents that keep working between conversations. Each one runs in its own browser in the cloud, so it can carry on with a task after you close ChatGPT. In late October, OpenAI let Pro users in the US message their Dots by iMessage or RCS.

The Catch

Login codes exist to prove a real person is signing in. Handing that step to an agent saves time, but it also means anyone who can steer the agent gets closer to your accounts. Websites can hide instructions in their pages to try to trick AI agents into doing things the user never asked for.

That is why the permission settings matter. The feature only works with an inbox the user has chosen to connect, and new sign-ins with saved logins still need a human to approve them. The next thing to watch is whether OpenAI extends the same trick to text message codes or bank checks, where the stakes are far higher.

Read Next: Samsung May Demand Trillion Won After Lam Leak Of Chip Secrets