Tenbbughunters

An issue exists in the conversation and contact filter APIs where user-supplied values in the `values` field of the filter payload are interpolated directly into SQL queries without parameterization when filtering by custom attributes of type date or number using the `is greater than` or `is less than` operators. This allows an authenticated user with account access to execute arbitrary SQL via time-based blind injection. Affected endpoints include '/api/v1/accounts/{account id}/conversations/filter', '/api/v1/accounts/{account id}/contacts/filter', and '/api/v1/accounts/{account id}/custom attribute definitions'.

Path traversal occurs in the Faiss and SimpleStore vector store implementations. The issue arises when the system accepts unsanitized `basePath` parameters from authenticated users. Attackers possessing valid API tokens can write vector store data to arbitrary locations on the filesystem, which may lead to data exfiltration or code execution.

The GraphCypherQAChain node forwards user-provided input directly into the Cypher query execution pipeline without proper sanitization. This allows an attacker to inject arbitrary Cypher commands that are executed on the underlying Neo4j database, potentially leading to data exfiltration, modification, deletion, or schema discovery. The issue occurs within the `run()` function of the `GraphCypherQAChain.ts` file, where the `query` variable is passed to the chain without escaping. Exploitation requires access to the prediction endpoint '/api/v1/prediction/{flowId}'.

Flowise has an issue where the NVIDIA NIM router endpoint ('/api/v1/nvidia-nim/*') was incorrectly whitelisted in the global authentication middleware. This allowed unauthenticated access to sensitive container management and token generation endpoints. Specifically, the issue stems from a configuration in `packages/server/src/utils/constants.ts` where the NVIDIA NIM route was added to the authentication whitelist, bypassing JWT/API-key validation. Affected endpoints include those for obtaining NVIDIA API tokens, managing container lifecycles (starting, stopping, pulling images), and retrieving container information. An attacker could exploit this to leak valid NVIDIA API tokens, manipulate container runtime, and potentially cause denial of service. The NVIDIA API token grants access to NVIDIA's inference API and can list over 170 large language models.

An information exposure issue exists in the 'POST /api/v1/account/forgot-password' endpoint. Unauthenticated attackers can enumerate valid email addresses and harvest sensitive user data, including PII (Personally Identifiable Information), user IDs, names, account status, and timestamps, by sending requests with known email addresses. This occurs because the endpoint returns full user objects.

Nginx UI contains a critical flaw where the `/api/backup` endpoint is accessible without authentication. When this endpoint is accessed via a GET request, the server provides a full system backup and discloses the AES-256 encryption keys (key and IV) required to decrypt the backup within the `X-Backup-Security` response header. This allows an unauthenticated remote attacker to download and immediately decrypt sensitive data, including user credentials, session tokens, SSL private keys, and Nginx configurations. The issue is rooted in the `CreateBackup()` function within `api/backup/router.go` and `api/backup/backup.go`, where authentication middleware is missing for the backup route. Approximately 500 active instances were identified in the RuNet segment, with 35% potentially affected.

Gradio is a Python package for rapid prototyping. Applications running outside of Hugging Face Spaces, versions 4.16.0 through 6.5.9, improperly handle OAuth components like `gr.LoginButton`. Visiting the `/login/huggingface` API endpoint triggers a process where the server retrieves and stores its Hugging Face access token in a session cookie. This cookie, signed with a hardcoded secret (`"-v4"`), can be decoded by remote attackers, potentially leading to credential theft. The vulnerable process occurs when an application is accessible over a network.

Gogs, a self-hosted Git service, has an issue where the DeleteComment API does not properly verify if a comment belongs to the repository specified in the URL. This allows a repository administrator to delete comments from any other repository by providing arbitrary comment IDs, effectively bypassing authorization controls. The `DeleteComment` function retrieves a comment by ID without validating repository ownership, and the `Database` function `DeleteCommentByID` does not perform repository validation. The vulnerable API endpoint is `/owner/repo/issues/comments/:id/delete`. The vulnerable parameter is `:id`, representing the comment ID.