OpenAI: Hardware Passkeys for Trusted AI Access

OpenAI will soon require hardware-backed passkeys for certain users. Starting September 1, 2026, members of its Trusted Access for Cyber program must use Advanced Account Security and a hardware-backed passkey to access frontier AI models. Users who don't meet this requirement will still be able to use ChatGPT, but their model access will return to a standard level. This program is for high-level security researchers needing access to restricted capabilities. To get trusted access, individuals must complete an identity verification process. This includes a government ID check, know-your-customer checks, and device health signals, all conducted via Persona. This helps OpenAI confirm the user is real and engaged in legitimate cybersecurity work. The system uses two identity controls: identity proofing to link a real person to the account, and a hardware-backed passkey to protect that verified account during subsequent logins. This physical key stores the private authentication credential, making it much harder for attackers to compromise. While Yubico has introduced custom YubiKeys, OpenAI says users can enroll with any compatible FIDO security key. This change aims to significantly enhance security for access to sensitive AI capabilities.