OpenAI rolls out advanced security mode for high-risk ChatGPT and Codex accounts

OpenAI is introducing a new account protection feature for users who face a higher risk of phishing, account takeover or targeted cyberattacks.

The company has launched Advanced Account Security, an optional setting for personal ChatGPT accounts that adds stricter sign-in requirements and stronger safeguards to reduce the risk of unauthorised access and data exposure. OpenAI says the feature is designed for people whose accounts may contain sensitive personal or professional information, including journalists, elected officials, political dissidents, researchers, cybersecurity professionals and other security-conscious users.

The move reflects the growing importance of AI accounts as high-value digital assets. As users increasingly rely on ChatGPT and Codex for work, research, coding and personal tasks, those accounts can accumulate sensitive context, business information, private conversations and connected workflow data. OpenAI said Advanced Account Security is intended to provide an additional layer of defence for users who may be more likely to be targeted.

Once enabled, the feature removes ordinary password-based access. Users must instead rely on phishing-resistant authentication, using two physical security keys, two passkeys, or one of each. OpenAI says users can use any FIDO-compatible security key, as well as passkeys.

The feature also removes email and SMS-based account recovery, which are common targets for attackers. Instead, recovery is limited to recovery keys, backup passkeys or physical security keys. That means users who enable the feature take on more responsibility for maintaining access to their own account.

OpenAI has also partnered with Yubico, the maker of YubiKey hardware security keys, to offer lower-cost YubiKey bundles to eligible Advanced Account Security users. The partnership is designed to make hardware-backed account protection more accessible for users who want stronger authentication.

A key part of the security model is that OpenAI support will no longer be able to recover an account once Advanced Account Security is enabled. While that creates a higher risk of permanent lockout if a user loses their recovery options, it also prevents attackers from attempting to compromise accounts by socially engineering support channels.

The feature also shortens sign-in windows and session duration, meaning users will be required to reauthenticate more often. It sends alerts when someone logs into the account and points users to a dashboard where they can review active ChatGPT and Codex sessions.

OpenAI has also made privacy protections stricter for users who enable the setting. While ChatGPT users can already opt out of having their conversations used to improve models, that exclusion is turned on by default for Advanced Account Security users.

The rollout comes as OpenAI expands its broader cybersecurity strategy. Earlier this year, the company launched its Trusted Access for Cyber program, which gives verified cybersecurity professionals and organisations access to advanced AI capabilities for legitimate defensive work. OpenAI later expanded that program with GPT-5.4-Cyber, a model designed for cyber defence tasks and made available through higher trusted-access tiers.

Members of OpenAI’s Trusted Access for Cyber program will be required to enable Advanced Account Security from June 1, unless they submit an alternative attestation showing they use phishing-resistant authentication through an enterprise single sign-on system.

The feature follows a wider trend among major technology companies toward stronger account security for high-risk users. Google has offered an Advanced Protection Program for years, built around phishing-resistant authentication and stricter recovery protections. OpenAI’s version applies a similar philosophy to AI accounts, which are becoming increasingly important repositories of personal and professional information.

For everyday users, Advanced Account Security may be more protection than necessary, particularly because losing access to recovery keys or security keys could make account recovery impossible. But for people handling sensitive work or facing targeted threats, the trade-off may be worth it.

The launch underlines a broader shift in how AI accounts are being viewed. They are no longer just login portals for a chatbot. For many users, they are becoming connected workspaces, coding environments, research assistants and stores of sensitive context. That makes account takeover a more serious risk, and makes stronger authentication an increasingly important part of using AI safely.

Photo Credit: DepositPhotos.com