CVE-2026-55443
MediumCVSS 5.1Exploitation Probability (EPSS)
Low risk5th percentile - higher than 5% of all known CVEs
Summary
LangChain prior to version 1.3.9 contains a vulnerability where several components that resolve filesystem paths or expand search patterns do not consistently confine the resolved path to the intended root directory. Affected behaviors include file-search middleware, configuration loaders, and path-prefix authorization checks that can allow access to files outside the configured root via glob patterns, symlinks, or path segment boundary bypass.
Risk Assessment
The risk involves potential disclosure of sensitive files outside the intended boundary when LangChain components receive path values, search patterns, or workspace contents from an untrusted source, including an LLM acting on untrusted input.
Recommendation
Immediately update LangChain to version 1.3.9 or later, which contains the fix for this vulnerability.
Related vulnerabilities
- CVE-2026-94603Critical
Podman from 4.4.0 until 5.8.8 and 6.1.3 treats an image carrying the io.podman.annotations.checkpoint.runtime.name annotation as a checkpoint and silently ignores user-supplied container creation restrictions such as --cap-drop=ALL. This allows image-controlled checkpoint state to select weaker sandbox settings and capabilities. Rootless containers are not affected because checkpoint and restore are unavailable in that mode. This issue is fixed in versions 5.8.8 and 6.1.3.
- CVE-2026-77459Critical
Argo CD from 2.10.0 until 3.3.15, 3.4.10, 3.5.4, and 3.6.0-rc2 creates PostDelete resource hooks without applying the AppProject destination and resource authorization checks used during ordinary synchronization. A user who can write to an Application's Git repository can commit a delete hook targeting a forbidden namespace, server, group, kind, or resource name, and deletion of the Application causes the application-controller to create that resource with its destination-cluster credentials. This issue is fixed in versions 3.3.15, 3.4.10, 3.5.4, and 3.6.0-rc2.
- CVE-2026-69227Critical
A flaw in Portal for ArcGIS allows unauthenticated access to certain /portaladmin/local API endpoints under specific conditions. This could enable an attacker to access administrative functionality that should require authentication.
- CVE-2026-55562Critical
Cronicle prior to 0.9.119 concatenates the attacker-controlled item.user.full_name value into HTML in the administrator Activity Log (htdocs/js/pages/admin/Activity.js) without applying encode_entities and inserts the result through this.div.html. The bundled pixl-server-user input filter can be bypassed by a tag containing a line break because its regular expression does not match newlines. An authenticated non-administrator can update their own name through /api/user/update, and when free_accounts is enabled an anonymous attacker can plant the name during self-registration. The stored name executes script in an administrator's session when the Activity Log is opened, enabling administrator takeover and subsequent command execution. This issue is fixed in version 0.9.119.
- CVE-2026-55183Critical
RClone Manager prior to 0.2.7 starts the local rclone rcd service on 127.0.0.1:51900 with --rc-no-auth, --rc-allow-origin *, and --rc-serve. A malicious web page opened by a user while the daemon is running can send cross-origin requests to the unauthenticated local API with the user's operating-system privileges. The requests can disclose configured storage credentials and potentially access, upload, modify, or exfiltrate locally stored files. The possible command-execution impact remains unconfirmed. This issue is fixed in version 0.2.7.
- CVE-2026-54454Critical
OpenCATS prior to 0.10.0, in the Import module (modules/import/ImportUI.php), can preserve an uploaded file's original name for one import data type and place the file in the web-accessible temporary directory. An authenticated user with Import privileges can select the affected import path and upload a PHP file that remains under an attacker-controlled executable filename, and an import error can leave the file on disk. If the temporary directory is served with PHP execution enabled, requesting the staged file can execute arbitrary PHP code with the web server process privileges. This issue is fixed in version 0.10.0.
- CVE-2026-110116Critical
Risesoft-y9 Digital-Infrastructure 9.6.5 through 9.6.11 contains a path traversal vulnerability in the fileManager storeFile endpoint that allows unauthenticated attackers to write arbitrary files. By sending crafted fullPath and fileName parameters to POST /rest/storeFile, attackers can create or overwrite service-writable files, potentially achieving code execution.
- CVE-2026-109412Critical
An exposed dangerous endpoint in Ivanti Neurons Patch for Intune allows a remote authenticated attacker to leak sensitive authentication information.
- CVE-2026-106042Critical
An External Control of File Name or Path vulnerability in Ivanti Xtraction before version 2026.4 allows a remote authenticated attacker to redirect application-generated requests to an attacker-controlled endpoint and obtain EPMM connector credentials transmitted by the application.
- CVE-2026-102989Critical
TanStack Router is a client-first, server-capable, fully type-safe router and full-stack framework for the web. From 1.143.12 until 1.169.39 for @tanstack/start-server-core, 1.168.60 for @tanstack/react-start, 1.168.57 for @tanstack/solid-start, and 1.168.56 for @tanstack/vue-start, an unauthenticated attacker can place non-public fields in a server-function request payload so that attacker-controlled middleware state reaches the error path and a response-shaped object is accepted as an HTTP response. A crafted server-function URL can therefore return attacker-controlled HTML from the application's origin; when a visitor opens the URL, the injected script can access same-origin data and issue requests with the visitor's authority.
Original NVD description (English source)
LangChain is a framework for building agents and LLM-powered applications. Prior to 1.3.9, several LangChain components that resolve filesystem paths or expand search patterns do not consistently confine the resolved path to the intended root directory. Affected behaviors include: a file-search agent middleware that validates a starting directory but not the search pattern or the resolved target of matched files, so glob patterns and symlinks can reach files outside the configured root; prompt- and chain/agent-configuration loaders that accept path fields and resolve them without confining the result to a trusted base or rejecting symlink targets; and path-prefix authorization checks that compare by string prefix without a path-segment boundary, so a sibling path sharing the prefix is accepted. When these components receive path values, search patterns, or workspace contents influenced by an untrusted source — including an LLM acting on untrusted input — the result can be disclosure of files outside the intended boundary. This vulnerability is fixed in 1.3.9.
Vulnerability data from NVD (NIST) · CISA KEV · EPSS

