Anythingllm

A vulnerability in mintplex-labs/anything-llm allows for a denial of service (DoS) condition through the modification of a user's `id` attribute to a value of 0. This issue affects the current version of the software, with the latest commit id `57984fa85c31988b2eff429adfc654c46e0c342a`. By exploiting this vulnerability, an attacker, with manager or admin privileges, can render a chosen account completely inaccessible. The application's mechanism for suspending accounts does not provide a means to reverse this condition through the UI, leading to uncontrolled resource consumption. The vulnerability is introduced due to the lack of input validation and sanitization in the user modification endpoint and the middleware's token validation logic. This issue has been addressed in version 1.0.0 of the software.Show less

A race condition vulnerability exists in the mintplex-labs/anything-llm repository, specifically within the user invite acceptance process. Attackers can exploit this vulnerability by sending multiple concurrent requests to accept a single user invite, allowing the creation of multiple user accounts from a single invite link intended for only one user. This bypasses the intended security mechanism that restricts invite acceptance to a single user, leading to unauthorized user creation without detection in the invite tab. The issue is due to the lack of validation for concurrent requests in the backend.Show less

In mintplex-labs/anything-llm, an attacker can exploit improper input validation by sending a malformed JSON payload to the '/system/enable-multi-user' endpoint. This triggers an error that is caught by a catch block, which in turn deletes all users and disables the 'multi_user_mode'. The vulnerability allows an attacker to remove all existing users and potentially create a new admin user without requiring a password, leading to unauthorized access and control over the application.Show less

mintplex-labs/anything-llm is vulnerable to improper input validation, allowing attackers to read and delete arbitrary files on the server. By manipulating the 'logo_filename' parameter in the 'system-preferences' API endpoint, an attacker can construct requests to read sensitive files or the application's '.env' file, and even delete files by setting the 'logo_filename' to the path of the target file and invoking the 'remove-logo' API endpoint. This vulnerability is due to the lack of proper sanitization of user-supplied input.Show less

mintplex-labs/anything-llm is vulnerable to a relative path traversal attack, allowing unauthorized attackers with a default role account to delete files and folders within the filesystem, including critical database files such as 'anythingllm.db'. The vulnerability stems from insufficient input validation and normalization in the handling of file and folder deletion requests. Successful exploitation results in the compromise of data integrity and availability.Show less

A mass assignment vulnerability exists in the `/api/invite/:code` endpoint of the mintplex-labs/anything-llm repository, allowing unauthorized creation of high-privileged accounts. By intercepting and modifying the HTTP request during the account creation process via an invitation link, an attacker can add a `role` property with `admin` value, thereby gaining administrative access. This issue arises due to the lack of property allowlisting and blocklisting, enabling the attacker to exploit the system and perform actions as an administrator.Show less

A stored Cross-Site Scripting (XSS) vulnerability exists in the chat functionality of the mintplex-labs/anything-llm repository, allowing attackers to execute arbitrary JavaScript in the context of a user's session. By manipulating the ChatBot responses, an attacker can inject malicious scripts to perform actions on behalf of the user, such as creating a new admin account or changing the user's password, leading to a complete takeover of the AnythingLLM application. The vulnerability stems from the improper sanitization of user and ChatBot input, specifically through the use of `dangerouslySetInnerHTML`. Successful exploitation requires convincing an admin to add a malicious LocalAI ChatBot to their AnythingLLM instance.Show less

A Denial of Service (DoS) vulnerability exists in the mintplex-labs/anything-llm repository when the application is running in 'just me' mode with a password. An attacker can exploit this vulnerability by making a request to the endpoint using the [validatedRequest] middleware with a specially crafted 'Authorization:' header. This vulnerability leads to uncontrolled resource consumption, causing a DoS condition.Show less

A vulnerability in mintplex-labs/anything-llm allows users with manager roles to escalate their privileges to admin roles through a mass assignment issue. The '/admin/system-preferences' API endpoint improperly authorizes manager-level users to modify the 'multi_user_mode' system variable, enabling them to access the '/api/system/enable-multi-user' endpoint and create a new admin user. This issue results from the endpoint accepting a full JSON object in the request body without proper validation of modifiable fields, leading to unauthorized modification of system settings and subsequent privilege escalation.Show less

In mintplex-labs/anything-llm, an improper input validation vulnerability allows attackers to escalate privileges by deactivating 'Multi-User Mode'. By sending a specially crafted curl request with the 'multi_user_mode' parameter set to false, an attacker can deactivate 'Multi-User Mode'. This action permits the creation of a new admin user without requiring a password, leading to unauthorized administrative access.Show less

mintplex-labs/anything-llm is vulnerable to path traversal attacks due to insufficient validation of user-supplied input in the logo filename functionality. Attackers can exploit this vulnerability by manipulating the logo filename to reference files outside of the restricted directory. This can lead to unauthorized reading or deletion of files by utilizing the `/api/system/upload-logo` and `/api/system/logo` endpoints. The issue stems from the lack of filtering or validation on the logo filename, allowing attackers to target sensitive files such as the application's database.Show less

As a default user on a multi-user instance of AnythingLLM, you could execute a call to the `/export-data` endpoint of the system and then unzip and read that export that would enable you do exfiltrate data of the system at that save state. This would require the attacked to be granted explicit access to the system, but they can do this at any role. Additionally, post-download, the data is deleted so no evidence would exist that the exfiltration occured.Show less

If an attacked was given access to an instance with the admin or manager role there is no backend authentication that would prevent the attacked from creating a new user with an `admin` role and then be able to use this new account to have elevated privileges on the instanceShow less

A user who is privileged already `manager` or `admin` can set their profile picture via the frontend API using a relative filepath to then user the PFP GET API to download any valid files. The attacker would have to have been granted privileged permissions to the system before executing this attack.Show less

Any user can delete an arbitrary folder (recursively) on a remote server due to bad input sanitization leading to path traversal. The attacker would need access to the server at some privilege level since this endpoint is protected and requires authorization.Show less

Enable exports of the database and associated exported information of the system via the default user role. The attacked would have to have been granted access to the system prior to the attack. It is worth noting that the deterministic nature of the export name is lower risk as the UI for exporting would start the download at the same time, which once downloaded - deletes the export from the system. The endpoint for exporting should simply be patched to a higher privilege level.Show less

Should an instance of AnythingLLM be hosted on an internal network and the attacked be explicitly granted a permission level of manager or admin, they could link-scrape internally resolving IPs of other services that are on the same network as AnythingLLM. This would require the attacker also be able to guess these internal IPs as `/*` ranging is not possible, but could be brute forced. There is a duty of care that other services on the same network would not be fully open and accessible via a simple CuRL with zero authentication as it is not possible to set headers or access via the link collector.Show less

A privilege escalation vulnerability exists in mintplex-labs/anything-llm, allowing users with 'default' role to delete documents uploaded by 'admin'. Despite the intended restriction that prevents 'default' role users from deleting admin-uploaded documents, an attacker can exploit this vulnerability by sending a crafted DELETE request to the /api/system/remove-document endpoint. This vulnerability is due to improper access control checks, enabling unauthorized document deletion and potentially leading to loss of data integrity.Show less

The inclusion of the web scraper for AnythingLLM means that any user with the proper authorization level (manager, admin, and when in single user) could put in the URL ``` http://169.254.169.254/latest/meta-data/identity-credentials/ec2/security-credentials/ec2-instance ``` which is a special IP and URL that resolves only when the request comes from within an EC2 instance. This would allow the user to see the connection/secret credentials for their specific instance and be able to manage it regardless of who deployed it. The user would have to have pre-existing knowledge of the hosting infra which the target instance is deployed on, but if sent - would resolve if on EC2 and the proper `iptable` or firewall rule is not configured for their setup.Show less