CVE-2026-54665

Langflow versions up to and including 1.6.9 contain a chained vulnerability that enables account takeover and remote code execution. An overly permissive CORS configuration (allow_origins='*' with allow_credentials=True) combined with a refresh token cookie configured as SameSite=None allows a malicious webpage to perform cross-origin requests that include credentials and successfully call the refresh endpoint. An attacker-controlled origin can therefore obtain fresh access_token / refresh_token pairs for a victim session. Obtained tokens permit access to authenticated endpoints — including built-in code-execution functionality — allowing the attacker to execute arbitrary code and achieve full system compromise.

The PDF reader in Mozilla Firefox before 39.0.3, Firefox ESR 38.x before 38.1.1, and Firefox OS before 2.2 allows remote attackers to bypass the Same Origin Policy, and read arbitrary files or gain privileges, via vectors involving crafted JavaScript code and a native setter, as exploited in the wild in August 2015.

In MLflow version 3.9.0, the MLflow Assistant feature introduced improper origin validation in its /ajax-api endpoints. This vulnerability allows a remote attacker to exploit cross-origin requests from a malicious webpage to interact with the MLflow Assistant running on a victim's local machine. By bypassing the loopback-only restriction, the attacker can modify the Assistant's configuration to enable full access, which in turn allows the execution of arbitrary commands via the Claude Code sub-agent. This issue is resolved in version 3.10.0.

A flaw was found in Keycloak. A remote attacker can exploit a Cross-Origin Resource Sharing (CORS) header injection vulnerability in Keycloak's User-Managed Access (UMA) token endpoint. This flaw occurs because the `azp` claim from a client-supplied JSON Web Token (JWT) is used to set the `Access-Control-Allow-Origin` header before the JWT signature is validated. When a specially crafted JWT with an attacker-controlled `azp` value is processed, this value is reflected as the CORS origin, even if the grant is later rejected. This can lead to the exposure of low-sensitivity information from authorization server error responses, weakening origin isolation, but only when a target client is misconfigured with `webOrigins: ["*"]`.

Impact: When using Socks5ProxyAgent, undici reuses a single connection pool across different origins without verifying that the pool's origin matches the requested origin. All requests are dispatched through the pool connected to the first origin, regardless of the intended destination. This causes cross-origin request routing: credentials and request data intended for origin B are sent to origin A, responses from the wrong origin are trusted, and HTTPS requests may be silently downgraded to HTTP. Impacted users are applications that use Socks5ProxyAgent (directly or via setGlobalDispatcher) and make requests to more than one origin. This was introduced in undici 7.23.0 via PR #4385 and affects all versions through 8.1.0. Patches: Upgrade to undici v7.26.0 or v8.2.0. Workarounds: Use a separate Socks5ProxyAgent instance per origin, or avoid using Socks5ProxyAgent with multiple origins.

Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.6, the chat message listener allows non-same-origin input:prompt and action:submit messages, so an external site can set prompt text and trigger submitPrompt() in an authenticated victim session. I validated this with a cross-origin attacker page that auto-posted messages and caused unauthorized POST /api/v1/chats/new and POST /api/chat/completions requests containing attacker-controlled prompts. This enables cross-site forced actions and model/tool execution under victim privileges without consent. This vulnerability is fixed in 0.9.6.