Security Vulnerabilities
- CVEs Published In March 2025
An arbitrary file deletion vulnerability exists in danny-avila/librechat version v0.7.5-rc2, specifically within the /api/files endpoint. This vulnerability arises from improper input validation, allowing path traversal techniques to delete arbitrary files on the server. Attackers can exploit this to bypass security mechanisms and delete files outside the intended directory, including critical system files, user data, or application resources. This vulnerability impacts the integrity and availability of the system.
In version 0.7.5 of danny-avila/LibreChat, there is an improper access control vulnerability. Users can share, use, and create prompts without being granted permission by the admin. This can break application logic and permissions, allowing unauthorized actions.
An improper access control vulnerability (IDOR) exists in the delete attachments functionality of danny-avila/librechat version v0.7.5-rc2. The endpoint does not verify whether the provided attachment ID belongs to the current user, allowing any authenticated user to delete attachments of other users.
Horovod versions up to and including v0.28.1 are vulnerable to unauthenticated remote code execution. The vulnerability is due to improper handling of base64-encoded data in the `ElasticRendezvousHandler`, a subclass of `KVStoreHandler`. Specifically, the `_put_value` method in `ElasticRendezvousHandler` calls `codec.loads_base64(value)`, which eventually invokes `cloudpickle.loads(decoded)`. This allows an attacker to send a malicious pickle object via a PUT request, leading to arbitrary code execution on the server.
A vulnerability in haotian-liu/llava v1.2.0 allows an attacker to cause a Denial of Service (DoS) by appending a large number of characters to the end of a multipart boundary in a file upload request. This causes the server to continuously process each character, rendering the application inaccessible.
A vulnerability in langgenius/dify versions <=v0.9.1 allows for code injection via internal SSRF requests in the Dify sandbox service. This vulnerability enables an attacker to execute arbitrary Python code with root privileges within the sandbox environment, potentially leading to the deletion of the entire sandbox service and causing irreversible damage.
HTTP Request Smuggling vulnerability in netease-youdao/qanything version 1.4.1 allows attackers to exploit inconsistencies in the interpretation of HTTP requests between a proxy and a server. This can lead to unauthorized access, bypassing security controls, session hijacking, data leakage, and potentially arbitrary code execution.
An information disclosure vulnerability exists in the latest version of transformeroptimus/superagi. An attacker can leak sensitive user information, including names, emails, and passwords, by attempting to register a new account with an email that is already in use. The server returns all information associated with the existing account. The vulnerable endpoint is located in the user registration functionality.
lunary-ai/lunary is vulnerable to broken access control in the latest version. An attacker can view the content of any dataset without any kind of authorization by sending a GET request to the /v1/datasets endpoint without a valid authorization token.
In lunary-ai/lunary v1.5.0, improper privilege management in the models.ts file allows users with viewer roles to modify models owned by others. The PATCH endpoint for models does not have appropriate privilege checks, enabling low-privilege users to update models they should not have access to modify. This vulnerability could lead to unauthorized changes in critical resources, affecting the integrity and reliability of the system.