ChatGPT passkeys come with a catch: lose your YubiKey, lose your account
OpenAI launched Advanced Account Security (AAS) on April 30, a set of opt-in protections that remove password-based login from ChatGPT and Codex accounts entirely. The feature replaces traditional credentials with FIDO2-compliant passkeys, software-based authenticators, or physical hardware keys. The tradeoff is stark: once enabled, OpenAI's support team cannot help recover a locked account.
The new security tier is recommended for journalists, researchers, political dissidents, and elected officials who use ChatGPT for sensitive work. OpenAI CISO Dane Stuckey said the company had already made YubiKeys standard for internal employee protection, and AAS extends that same phishing-resistant model to users. A partnership with Yubico delivers a custom two-pack of co-branded keys at a discounted price: the YubiKey C Nano, designed to stay plugged into a laptop port for daily use, and the YubiKey C NFC for backup and cross-device authentication across laptops and mobile devices.
AAS also shortens sign-in sessions and alerts users about new logins with session management controls. Account recovery shifts from email and SMS to backup passkeys and recovery keys.
Users who enroll automatically receive training exclusion, meaning their conversations will not be used to train AI models. The partnership goes beyond login security. Yubico acting CEO Jerrod Chong said the intent is to "drastically reduce the threat of unauthorized access to sensitive data in OpenAI accounts worldwide." But Yubico's longer play targets agentic AI workflows where machines act autonomously.
Through its Role Delegation Token (RDT) architecture, Yubico is building a cryptographic, hardware-rooted authorization chain that requires a physical YubiKey tap to sign off on high-consequence AI actions. If the execution context changes between approval and execution, the system blocks the action and forces reauthorization. The collaboration marks the first time a major AI company has tied hardware-backed authentication to both account login and agentic action authorization. As enterprises push toward autonomous AI workflows, the ability to produce hardware-attested proof that a human explicitly approved a specific action could become a compliance requirement.
Yubico has also partnered with Delinea to extend hardware-attested proof of human authorization into audit trails that bind every automated action to a verified human.
