NocoDB versions before 2026.10.1 have an information disclosure vulnerability that allows unauthenticated users holding shared view links to access hidden columns. By exploiting the public group-by endpoint with specific parameters, attackers can reveal hidden column values, counts, and existence. This vulnerability is fixed in version 2026.10.1.
NocoDB before 2026.10.0 contains an account takeover vulnerability that allows users with the base creator role to hijack other accounts by resending base invitations. Attackers can trigger baseUserInviteResend to mint a new invite_token, read it from the base member list, and reset the victim's password through signup, taking over owner or super admin accounts.
NocoDB before 2026.10.0 contains an information disclosure vulnerability that allows unauthenticated attackers to read hidden columns by supplying comma-separated group-by column lists. Attackers with a shared view link can pair a visible and hidden column to bypass restrictSharedViewColumnReferences and retrieve hidden values tied to individual rows.
NocoDB versions before 2026.10.1 have an SQL injection vulnerability due to improper escaping of backslashes in select-field option titles embedded in MySQL or MariaDB enum/set column definitions. This flaw allows a base creator or owner to manipulate option titles to inject malicious ALTER TABLE commands, potentially modifying database schema elements such as columns and indexes.
NocoDB before 2026.10.1 contains a consent phishing vulnerability that allows unauthenticated attackers to register OAuth clients with arbitrary HTTPS redirect URIs through the unauthenticated dynamic client registration endpoint. Attackers can impersonate trusted applications on a consent screen lacking unverified-client warnings or redirect host display, obtaining tokens to read and write victim tables and records.
NocoDB from 0.301.4 before 2026.10.0 contains an information disclosure vulnerability that allows authenticated workspace viewers to read pending invitation tokens through the workspace member list. Attackers can redeem an exposed higher-privileged invitation token at signup to claim the invited account and escalate to roles such as workspace owner.
NocoDB before 2026.10.1 contains a credential exposure vulnerability in the connection-test endpoint that allows authorized integration users to redirect stored credentials by overriding connection host, port, and user fields. Attackers can point the test at a server they control to capture offline-crackable PostgreSQL md5 or MySQL handshakes and recover the stored database password.
NocoDB versions before 2026.10.1 have an authentication bypass vulnerability due to insufficient session expiration. Attackers holding revoked OAuth access tokens can reuse them via the session header to authenticate as first-party sessions. This bypasses token revocation, client existence checks, route allowlists, and granted resource scopes, enabling unauthorized profile modifications, storage uploads, and forced global sign-outs.
NocoDB from 2026.07.0 before 2026.10.0 contains an sql injection vulnerability that allows users with Creator role to inject SQL through unescaped backslashes in the DATETIME_FORMAT formula format argument on MySQL-backed bases. Attackers can place a backslash before a quote to terminate the DATE_FORMAT string literal and read any data reachable by the data source's database user, including tables outside the base.
NocoDB from 0.253.0 before 2026.09.1 contains an incorrect authorization vulnerability in baseUserDelete that allows base Creators to remove higher-privileged Owners due to a broken role power comparison. Attackers holding the Creator role can remove other Owners of a base, since getProjectRolePower returns -1 for the target and never refuses the request.
NocoDB before 2026.10.1 contains a template injection vulnerability that allows authenticated Creator-role users to execute server-side JavaScript by supplying array-form property paths to Handlebars helpers in webhook templates. Attackers can bypass the string-only prototype guard to reach the function constructor, running operating-system commands with server privileges or exfiltrating process-internal values via outbound requests.
NocoDB versions before 2026.10.1 have an information disclosure vulnerability due to predictable file paths for data exports. Unauthenticated attackers can reconstruct these paths using table and view metadata to download full data exports, including fields hidden from public views.
NocoDB before 2026.10.1 contains an incorrect authorization vulnerability that allows organization Creators to reach routes restricted to other roles because the aclFn middleware never enforced allowedRoles. Attackers can generate a super admin password reset link via org-users endpoints to take over the instance, or read and modify SSO client secrets.
NocoDB before 2026.10.1 contains a stored HTML injection vulnerability that allows authenticated users with record edit rights to inject forms, images, and styled overlays into rich text cells, which are sanitized only with DOMPurify's default policy. Attackers can place a page-sized spoofed sign-in form or remote image in a cell to harvest passwords and disclose viewer IP addresses in expanded records, gallery and kanban cards, and shared views.
NocoDB before 2026.10.0 contains an sql injection vulnerability that allows authenticated users with the Creator role to inject SQL by supplying a crafted formula parsed_tree during table creation. Attackers can set a malicious binary-expression operator that is interpolated into a knex.raw fragment, reading user and token tables and escalating to super administrator on SQLite installs.
NocoDB versions prior to 2026.10.1 contain a stored CSS injection vulnerability that allows authenticated users with Commenter or Editor roles to inject inline styles via single quotes in mention text. This injection can manipulate the user interface for other viewers by breaking out of the single-quoted data-id attribute in comments or rich text cells.
vLLM from 0.6.0 before 0.33.0 lacks a request-wide media budget, allowing remote inference clients to exhaust frontend resources by submitting many remote image URLs in one request. Attackers can send a compact request to /v1/messages or /v1/responses that forces downloading and decoding of every image before prompt-length admission, exhausting memory, CPU, and connections.
vLLM before 0.33.0 contains a path traversal vulnerability that allows malicious model publishers to redirect the served model and tokenizer by setting speculators_config verifier.name_or_path in config.json. Attackers can point this field at another local checkpoint directory to leak its tokenizer vocabulary via unauthenticated /detokenize requests and gain query access to private models.
vLLM from 0.28.0 before 0.33.0 contains an improper input validation vulnerability that allows unauthenticated attackers to crash the shared EngineCore by sending out-of-range prompt token ids when served with --skip-tokenizer-init. Attackers can submit a large positive token id to /v1/completions, /v1/embeddings, or /pooling, bypassing the tokenizer-gated vocab check and causing a fatal embedding index error that denies service to all tenants.
CVE-2026-130683 is a critical path traversal vulnerability in siyuan-note versions before 3.8.10. The application uses an incomplete deny-list for sensitive paths, omitting important shell initialization files and directories. Authenticated administrators can exploit this flaw by calling the /api/file/copyFile endpoint with an absolute path outside the workspace, allowing them to write malicious payloads that execute with user privileges upon the next shell or login. This vulnerability is fixed in version 3.8.10.