CVE-2026-55423: 
Homebrew vulnerability analysis and mitigation

CVE-2026-55423 is an Insufficient Session Expiration vulnerability in Langflow, a tool for building and deploying AI-powered agents and workflows. The logout button fails to properly clear the user session, leaving the previous user authenticated unless another user explicitly logs in. All Langflow versions prior to 1.7.0 are affected. The vulnerability was published on June 19, 2026, and carries a CVSS v3.1 base score of 6.1 (Medium) with a physical attack vector (Github Advisory, Langflow Advisory).

The root cause is a dual-layer failure classified as CWE-613 (Insufficient Session Expiration). On the backend, the /logout endpoint deleted authentication cookies (access_token_lf, refresh_token_lf, apikey_tkn_lflw) without matching the original httponly, samesite, secure, and domain parameters used when the cookies were created; browsers require exact parameter matching to honor cookie deletion, so the cookies persisted. On the frontend, the auth store did not call any cookie-clearing logic on logout, leaving access_token_lf present in both Local Storage and Cookies and refresh_token_lf present in Cookies. This issue manifests specifically when LANGFLOW_AUTO_LOGIN is set to False (Langflow Advisory, PR #10527, PR #10528).

A person with physical access to a shared workstation running Langflow can refresh the browser after the previous user clicks logout and regain full access to that user's authenticated session — without knowing any credentials. This grants high confidentiality and integrity impact: the attacker can read all data accessible to the prior user's account (AI workflows, API keys, configurations) and modify or delete it. Availability is not directly impacted, but unauthorized modification of AI workflows or agent configurations could have downstream operational consequences (Langflow Advisory, Github Advisory).

A proof-of-concept is publicly documented in the official security advisory: clicking logout and then refreshing the page restores the previous user's session. The EPSS score is approximately 0.15%, indicating low predicted exploitation probability in the near term. There is no evidence of in-the-wild exploitation, no known threat actor attribution, and the vulnerability is not listed in the CISA KEV catalog. Exploitation requires physical or local access to the victim's browser session, significantly limiting the attack surface (Langflow Advisory, Github Advisory).

1. Identify a shared Langflow instance: Locate a Langflow deployment (version < 1.7.0) running on a shared computer or kiosk, with LANGFLOW_AUTO_LOGIN set to False.
2. Wait for the legitimate user to log out: Observe the legitimate user click the Logout button in the Langflow UI, which appears to terminate the session.
3. Refresh the browser: Press F5 or navigate back to the Langflow URL. Because the access_token_lf and refresh_token_lf cookies were not properly cleared, the browser resubmits them and the application restores the previous user's authenticated session.
4. Access the victim's account: The attacker now has full access to the previous user's Langflow workspace, including AI workflows, API keys, agent configurations, and any data accessible to that account — without entering any credentials (Langflow Advisory).

• Browser Storage: Presence of access_token_lf in browser Local Storage and Cookies after a logout action; presence of refresh_token_lf in Cookies after logout.
• Cookies: LANGFLOW_AUTO_LOGIN_OPTION cookie persisting in the browser after the user has clicked logout.
• Logs: Server-side access logs showing authenticated API requests (using a previously issued token) immediately following a /logout endpoint call from the same session or IP, indicating the session was not invalidated.
• Application Behavior: Users reporting they are returned to an authenticated dashboard after logging out and refreshing the page (Langflow Advisory, PR #10527).

Upgrade Langflow to version 1.7.0 or later, which includes fixes from PRs #10527 and #10528. PR #10527 corrects the backend /logout endpoint to delete cookies using the same httponly, samesite, secure, and domain parameters used at login. PR #10528 updates the frontend cookieManager.clearAuthCookies() to also remove the LANGFLOW_AUTO_LOGIN_OPTION cookie and integrates this call into the auth store's logout flow. As an interim workaround, restrict physical access to workstations running Langflow and instruct users to manually clear browser cookies and Local Storage after logging out (Github Advisory, PR #10527, PR #10528).