/

CVE-2026-33757

Severity:

CRITICAL RISK

(9.6)

Session Management and Authentication Logic Flaw

CVE-2026-33757

Modified

Vulnerability type

Category:

Session Management and Authentication Logic Flaw

CWE:

CWE-384

Impact:

Session Fixation and Account Compromise

Severity:

CRITICAL RISK

(9.6)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

OpenBao's JWT/OIDC login flow using roles with callback_mode=direct can complete authentication without explicit user confirmation, enabling a form of remote phishing.

An attacker who starts an OIDC authorization request can trick a victim into visiting a crafted URL that silently binds the victim's browser to the attacker-initiated session, resulting in a session fixation scenario.

Affected context

This vulnerability arises in OpenBao deployments that:

  • Use JWT/OIDC authentication
  • Have one or more roles configured with callback_mode=direct
  • Rely on the OpenBao side to drive the OIDC authorization code flow

In this context, the OIDC callback is made directly to the OpenBao API, and the login can complete automatically once the user follows the crafted URL, without an intermediate confirmation or explicit user intent validation.

Exploit preconditions

  • Attack surface: Web-based authentication endpoints for JWT/OIDC login in OpenBao.
  • Required access:
  • Ability to initiate an OIDC authorization request against OpenBao using a role with callback_mode=direct.
  • Ability to deliver a crafted link or redirect to the victim (e.g., email, chat, malicious page).
  • Trigger point: The victim's browser follows the attacker-controlled OIDC login URL, causing OpenBao to finalize the authentication and associate the session with the attacker-initiated flow.
  • Why exploitation is possible: The direct callback mode completes the authorization code flow server-side and issues tokens without requiring a distinct, user-driven confirmation step to bind the session to the user's explicit intent, allowing session fixation or "remote phishing" of the login.

Why this happens

The core issue is a session management and authentication-flow design flaw mapped to CWE-384: Session Fixation.

In the callback_mode=direct configuration, OpenBao's JWT/OIDC integration completes the authorization code flow directly against the API and permits token retrieval via polling. Because the flow does not enforce an explicit confirmation screen or equivalent user interaction at the moment of session binding, an attacker can pre-initiate an authentication request and later cause an unsuspecting user to follow a URL that finalizes the attacker's session in the victim's browser context. This pattern effectively "fixes" the resulting authenticated session to parameters chosen by the attacker, rather than deriving it solely from a user-driven login initiation, which breaks standard guarantees of OIDC-based SSO flows.

Potential impact

  • Attacker-controlled session becoming active in the victim's browser.
  • Unauthorized access to resources managed by OpenBao under a session the victim did not intentionally initiate.
  • Potential escalation to broader account or secret access if that OpenBao session has high privileges.
  • User confusion and trust erosion due to unexpected or invisible logins.
  • Facilitation of more convincing phishing campaigns that appear to come from a legitimate OpenBao/OIDC login flow.

General mitigation guidance

Defensive strategies focus on strengthening session binding and user intent verification in authentication flows:

  • Require an explicit confirmation step (such as a confirmation screen or consent page) before finalizing OIDC/JWT logins, especially for flows where callbacks are handled directly by the backend.
  • Avoid configurations that allow "automatic" or silent login completions without a user action that is clearly tied to the specific session (e.g., clicking a clearly labeled "Continue" button after redirection).
  • Limit or remove use of roles that employ direct callback patterns unless strong compensating controls exist to verify user presence and intent.
  • Implement defense-in-depth by enforcing confirmation requirements or step-up authentication on the identity provider / token issuer side for client IDs used by OpenBao.
  • Follow least-privilege principles so that any session established via automated flows has minimal access until additional interactive verification is performed.

How teams detect issues like this

Detection should center on identifying risky authentication-flow configurations and anomalous login patterns:

  • Use configuration review or policy-as-code checks to flag roles with callback_mode=direct or similar "non-interactive" authorization flows.
  • Apply static analysis or rules-based scanning to IaC, config files, and deployment manifests that define OIDC/JWT roles, looking for modes that bypass explicit user confirmation.
  • Monitor authentication logs for:
  • Repeated authorization requests from attacker-controlled IPs followed by victim IPs completing logins.
  • Short-lived or unusual login sequences where the user is not expected to have initiated an OpenBao login.
  • Integrate automated security review tools into CI/CD to enforce that new or modified authentication configurations include explicit user interaction steps before session establishment.

Sources

SourceLink
datatracker.ietf.orgdatatracker.ietf.org/doc/html/rfc8628#section-5.4
github.comgithub.com/openbao/openbao/commit/e32103951925723e9787e33886ab6b6ec20f4964
github.comgithub.com/openbao/openbao/security/advisories/GHSA-7q7g-x6vg-xpc3
access.redhat.comaccess.redhat.com/security/cve/CVE-2026-33757
bugzilla.redhat.combugzilla.redhat.com/show_bug.cgi?id=2452269
security.access.redhat.comsecurity.access.redhat.com/data/csaf/v2/vex/2026/cve-2026-33757.json

Frequently asked

questions

Yes. An attacker can exploit this remotely by initiating an OIDC login flow using a callback_mode=direct role and then convincing the victim to open a crafted URL, which can silently finalize the attacker's session in the victim's browser context.

The attacker needs the ability to initiate JWT/OIDC authentication against the OpenBao instance using a role configured with callback_mode=direct, plus any common capability to send links or redirects to the victim (such as email, chat, or web content under their control).

The safest pattern is to enforce an explicit user confirmation step before completing any authentication flow, particularly for direct callback modes. This can be done by adding a confirmation screen on the OpenBao side or ensuring the identity provider requires user confirmation for each session associated with OpenBao's client ID.

Design authentication flows so that session establishment always depends on clear, user-driven actions at the time of login, not solely on backend callbacks. Regularly review authentication configurations for silent or automatic completion paths, apply principle-of-least-privilege to any sessions established by such flows, and incorporate configuration linting and security reviews into your CI/CD pipeline for identity and access configurations.

Find out what's already
exploitable in your codebase.