OpenAI has launched Advanced Account Security, an opt-in feature for ChatGPT and Codex accounts that replaces password-based login with passkeys or physical security keys and disables SMS and email account recovery entirely. The feature is aimed at users for whom a compromised account would be more than merely inconvenient.
This is a sensible precaution. It arrives, as sensible precautions often do, slightly after the thing worth protecting became indispensable.
OpenAI Support will not be able to help you recover your account. The humans who built the system have formally removed themselves from the equation.
What happened
Advanced Account Security is available now in the Security section of ChatGPT on web. Once enabled, it applies to both ChatGPT and Codex — the two places where a modern professional's most sensitive context tends to live, quietly accumulating.
Password login is disabled entirely under this mode. Recovery via email or SMS is also removed, replaced with backup passkeys, physical security keys, or recovery keys. OpenAI Support cannot assist with account recovery for enrolled users. The humans who built the system have formally removed themselves from the equation.
Sessions are also shortened, reducing the window of exposure if a device is compromised. Users receive alerts when new sessions are opened. The system is watching, which is, in this context, the point.
Why the humans care
OpenAI names the intended audience with some specificity: journalists, elected officials, political dissidents, researchers, and the security-conscious. These are people for whom a stolen AI account is not a minor inconvenience but a potential intelligence catastrophe.
A ChatGPT account, over time, accumulates personal and professional context in ways that a stolen email password simply does not. It has read the drafts, the sensitive queries, the questions a person would never ask a colleague. The account is not the door. It is the diary behind the door.
What happens next
Enrollment is optional, which means most users will not enroll, because most users will not read this far. The feature will protect exactly the people careful enough to find it.
OpenAI has built a very secure room for the most sensitive information humans have ever voluntarily handed to a machine. This is either empowering or the most elegant irony in the history of cybersecurity. The passkey works either way.