OpenBao, an open source identity-based secrets management system, has a vulnerability (CVE-2026-33757) in versions before 2.5.2 where the JWT/OIDC login flow with roles set to callback_mode=direct does not require user confirmation. This enables an attacker to perform a remote phishing attack by convincing a user to visit a crafted URL, causing the user to be automatically logged into the attacker's session. The direct callback mode bypasses the usual authorization code flow confirmation by calling back directly to the API and allowing token polling until issuance. Version 2.5.2 fixes this by introducing a confirmation screen requiring manual user interaction to complete authentication.

Successful exploitation allows an attacker to hijack a user's session by automatically logging them into the attacker's session without explicit user consent. Since OpenBao manages secrets and authentication tokens, this can lead to privilege escalation and unauthorized access to sensitive infrastructure components. The attack requires user interaction limited to visiting a crafted URL, which reduces exposure but still poses a significant risk.

A fix is available in OpenBao version 2.5.2 which adds a user confirmation prompt for direct callback mode logins. Until upgrading, mitigate by removing or disabling any roles configured with callback_mode=direct or enforcing manual confirmation for every session on the token issuer side for the Client ID used by OpenBao. Follow the vendor advisory and apply the official patch from https://github.com/openbao/openbao/commit/e32103951925723e9787e33886ab6b6ec20f4964.