CVE Feed

Adminer before 6.1.1 contains a remote code execution vulnerability that allows authenticated SQLite users to bypass ATTACH and VACUUM INTO restrictions by prefixing statements with a UTF-8 byte-order mark. Attackers can attach a database file at a PHP-served path and create a table named with PHP code to execute it as the web server user.

better-auth from 1.3.8-beta.3 before 1.7.0-rc.3 contains a UI misrepresentation vulnerability in the deviceAuthorization plugin because GET /device omits the requesting client_id and scope. Attackers can start a device request via /device/code, trick signed-in users into approving the code, then redeem it at /device/token for a live session token.

DOMPurify before 3.4.16 contains a cross-site scripting vulnerability that allows attackers to bypass sanitization because IN_PLACE mode returns a force-removed style element whose text content still carries markup. Attackers can supply a style element containing a closing style tag and an img onerror payload, which executes script when the application serializes and re-parses the returned node.

DOMPurify from 3.4.13 before 3.4.16 contains a cross-site scripting vulnerability because IN_PLACE mode fails to neutralize subtrees detached by afterSanitizeElements or afterSanitizeAttributes hooks. Attackers can supply markup with on* handlers inside a wrapper removed by such hooks, leaving handlers armed on the live DOM to execute script in the victim origin.

SurrealDB before 3.3.0 allows cross-tenant access because the SurrealQL USE statement switches a session to any namespace/database without checking that the caller's authenticated level covers it. Authenticated namespace, database, or record users can switch to another tenant and read, create, update, or delete rows in tables whose PERMISSIONS allow it, such as PERMISSIONS FULL.

SurrealDB from 3.0.0 before 3.3.0 contains a missing authentication vulnerability in the revoke RPC method that allows low-privileged or unauthenticated callers to invalidate other users' refresh tokens without proof of possession. Attackers can obtain grant identifiers via ACCESS SHOW ALL with VIEWER permission or leaked tokens, then permanently revoke refresh tokens to force re-authentication.

SurrealDB before 3.3.0 contains an authorization bypass vulnerability that allows callers with table-level write permission to read SELECT-restricted fields by resolving JSON Patch pointers against the unreduced record. Attackers can use PATCH copy or move operations to copy hidden fields into readable destinations, or abuse test mismatch errors as a disclosure oracle.

SurrealDB from 3.1.0 before 3.3.0 contains an authorization bypass vulnerability that allows authenticated users to delete protected graph edges by triggering a REFERENCE ON DELETE CASCADE. Attackers with delete permission on a referenced table can delete a record whose cascade disables permission checks for edge cleanup, erasing edges despite the edge table's PERMISSIONS FOR delete clause.

SurrealDB 3.1.0 before 3.3.0 lets Surrealism modules exhaust server memory because module stdout/stderr lines are buffered unbounded in host memory, bypassing the configured module memory limit. An attacker can run an Owner-installed malicious module that writes output without newlines, crashing the server and denying service to all namespaces and databases.

SurrealDB before 3.3.0 contains a denial of service vulnerability that allows query-capable users to crash the server by reading non-ASCII header values in embedded JavaScript. Attackers with scripting enabled can read such values via Headers get(), entries(), values(), or getSetCookie(), or return them from fetch() endpoints, aborting the server process.

SurrealDB before 3.3.0 contains an authorization bypass vulnerability that allows record users to infer field values hidden by field-level SELECT permissions using LIVE SELECT WHERE predicates over $before or $after. Attackers can craft repeated predicates and observe whether notifications are delivered on CREATE, UPDATE, or DELETE to confirm or narrow restricted field values.

SurrealDB before 3.3.0 contains an authorization bypass vulnerability that allows record users to read fields hidden by field-level SELECT permissions through LIVE SELECT projections. Attackers can project restricted fields from the unreduced $after or $before parameters under an alias to receive exact stored values in CREATE, UPDATE, and DELETE notifications.

SurrealDB from 3.0.0 before 3.3.0 contains an information disclosure vulnerability that allows record users to read values hidden by field-level SELECT permissions through COMPUTED fields. Attackers can select records whose readable COMPUTED fields derive from hidden fields, or use write RETURN clauses with $before/$after to expose hidden values.

SurrealDB before 3.3.0 does not check the caller-selected namespace or database against the level at which a namespace- or database-level system user authenticated, allowing cross-tenant access. An authenticated attacker can set the Surreal-NS/Surreal-DB headers on /sql or call the RPC use method to read, create, modify, and delete rows and read DEFINE PARAM values in another tenant where table PERMISSIONS allow it.

SurrealDB from 3.0.0 before 3.3.0 contains an authorization bypass vulnerability that allows authenticated record users to read records in other namespaces or databases by resolving record ids through FETCH. Attackers can switch to a foreign namespace/database and name a known record id in a RETURN … FETCH query to read another tenant's data when table PERMISSIONS allow reads.

Network-AI (npm network-ai) before 5.15.2 contains an origin validation error in McpSseServer._handleRequest in lib/mcp-transport-sse.ts that reflects localhost Origins into Access-Control-Allow-Origin regardless of bind interface. When the server binds a non-loopback address, an attacker page with a localhost Origin visited by a victim can read unauthenticated endpoints like /health cross-origin.

Network-AI from 5.14.0 before 5.15.2 contains a security control bypass vulnerability in the ClaudeHookBridge PreToolUse hook, which matches deny patterns only against the first present candidate field. A prompt-injected agent can place dangerous payloads in fields like Write content or an MCP script parameter beside a benign file_path or command, evading denyPatterns to write web shells or execute scripts.

Network-AI 5.0.0 before 5.15.2 contains a cross-site WebSocket hijacking vulnerability in DashboardServer that allows remote attackers to read live agent topology because handleUpgrade performs no Origin or token validation. Attackers can lure a victim running the dashboard to a malicious page that opens a WebSocket to the loopback dashboard and exfiltrates agent IDs, roles, task descriptions, and token budgets.

Admidio 4.3.2 through 4.3.17 and 5.0.0 before 5.0.15 contains a broken access control vulnerability that allows role administrators to assign users to roles in other organizations. Attackers can submit crafted assign requests to members_assignment.php with a foreign role UUID to gain privileged membership, log into that organization, and edit its user profiles.

Admidio 5.0.0 before 5.0.15 contains an unauthenticated stored cross-site scripting vulnerability in RegistrationPresenter.php, which inserts the raw FIRST_NAME registration value into an unencoded data-message attribute. Attackers can register with a double-quoted event handler that executes JavaScript in an administrator's session when they hover the entry on the Manage registrations page.

Admidio 4.0.12 through 4.3.17 and 5.0.0 before 5.0.15 contains a reflected cross-site scripting vulnerability in system/redirect.php that allows unauthenticated attackers to inject JavaScript via the url GET parameter, due to an incomplete fix for CVE-2021-43810. Attackers can craft a javascript: URL containing a decoy http:// segment that passes validation and executes script in the Admidio origin when victims click the redirect link.

Grav before 2.1.7 contains an information disclosure vulnerability because bundled nginx, lighttpd, IIS, and Caddy 2 configurations block .git only at document root. Unauthenticated attackers can request nested repositories such as user/.git to retrieve Git config, refs, and objects, reconstructing unpublished history or committed secrets.

Grav from 2.0.22 before 2.1.7 contains an authentication bypass vulnerability in the Clockwork profiler that trusts loopback REMOTE_ADDR values. Unauthenticated attackers behind a same-host HTTP proxy can read profiler records exposing system and plugin configuration, secrets, request metadata, and queries.

Grav before 2.1.7 contains a denial of service vulnerability that allows authenticated media authors to exhaust memory by uploading compressed oversized raster images. Attackers can reference a Twig or Markdown transform so GD or Imagick decodes the full source without dimension checks, letting anonymous requests repeatedly cause page unavailability.

Grav Star Ratings 2.0.1 through 2.0.3 contains a missing authorization vulnerability that allows unauthenticated visitors to vote by bypassing client-only vote_access and readonly enforcement. Attackers can use the anonymous nonce emitted by getData() to submit votes directly to addVote(), manipulating restricted or disabled ratings and aggregate rating output.

Grav Guestbook plugin before 0.5.2 contains a stored cross-site scripting vulnerability because its AJAX Load more and Admin Load all handlers insert stored author and message fields into the DOM as unescaped HTML via jQuery. Unauthenticated visitors can submit markup that executes in other visitors' browsers and, in versions 0.5.0 and 0.5.1, in administrator moderation sessions.

Grav Login OAuth plugin (grav-plugin-login-oauth) before 1.4.3 provisions OAuth accounts as ordinary Grav users with a deterministic <provider>.<id> username and an unsalted MD5 hash of the provider identifier as the password, without disabling password login. Unauthenticated remote attackers who learn or guess a provider identifier can reproduce these credentials and log in, gaining administrative control of privileged accounts.

The Grav API plugin (grav-plugin-api) before 1.0.36 contains an improper session validation vulnerability that allows disabled accounts to keep API access, because SessionAuthenticator does not enforce account state. Disabled users can reuse an already authorized browser or remember-me session to call API endpoints until expiry, defeating operator revocation.

Grav Guestbook plugin 0.5.0 and 0.5.1 contains missing authorization and cross-site request forgery vulnerabilities in Admin moderation actions delete, approve, and approveAll. Authenticated Admin users lacking the admin.guestbook permission, or external sites abusing a moderator's session via nonce-less GET requests, can delete or approve Guestbook entries.

Grav Git Sync 1.0.2 through 3.4.12 contains a command injection vulnerability that allows authenticated content editors to execute OS commands by injecting shell substitutions into profile fields. When the gravuser or gravfull author mode is enabled, attackers can set a malicious username, full name, or email that reaches PHP exec() via git commit --author, running commands as the web-server account.

Grav grav-plugin-api before 1.0.36 contains a privilege escalation vulnerability that allows delegated user managers to obtain super-admin access by storing positive scalar admin or api parent grants. Attackers with api.users.write can edit account access maps or invitations so the inheriting permission resolver satisfies requireSuper checks, accessing super-only API endpoints.

Grav Data Manager plugin 1.4.0 through 1.4.5 contains a missing authorization vulnerability that allows unauthenticated attackers to delete stored data files via the classic Admin integration. Attackers can send a request with the delete parameter and known data type and filename to remove files under user/data/, including form submissions, or forge requests lacking CSRF nonce checks.

Grav from 2.0.22 before 2.1.7 contains an information disclosure vulnerability that allows page editors to read the Clockwork debugger token because the sandbox redaction list omits system.debugger.token. Attackers can render the token through editor-authored Twig and supply it as X-Clockwork-Auth to read profiler records with configuration, request metadata, and database queries.

Grav Login plugin 1.2.0 before 3.9.9 contains an information disclosure vulnerability that allows publishers to obtain password-reset tokens by authoring a page at the reset route. When a victim opens an emailed reset link, the publisher-controlled page rendered by Login::getPage() exposes the username and bearer token, enabling takeover of more privileged accounts.

Grav Login plugin 2.7.0 before 3.9.9 contains an authentication bypass vulnerability that allows attackers with a valid password to skip two-factor authentication through remember-me cookie restoration. Attackers can obtain a persistent remember-me cookie after the password step and restore it as a fully authorized session without completing the second factor.

Grav Login plugin 2.7.0 before 3.9.9 contains an authentication bypass in processUserProfile(), which checks that the session user exists but not that a pending front-end 2FA session is fully authorized. Attackers with a victim's password who can publish a page form without an access rule can invoke update_user before completing 2FA to disable twofa_enabled and take over the account.

vLLM before 0.29.0 passes uploaded audio to the PyAV/FFmpeg av.open() fallback with only a file-size check, exposing FFmpeg's IAMF parser heap overflow. Authenticated remote attackers can repeatedly upload malformed IAMF containers to /v1/audio/transcriptions, corrupting the heap and crashing the API process.

SiYuan v3.1.12 before v3.8.6 fails to restrict the Sprig template function map in the /api/icon/getDynamicIcon content parameter, allowing resource exhaustion. A publish reader, or anyone when publish auth is disabled, can use uncapped functions like until to OOM-kill the kernel with one request.

SiYuan v3.7.0 before v3.8.6 contains an absolute path traversal vulnerability in the kernel's private plugin HTTP server, which passes a kernel plugin handler's ResponseFile.Path to gin's Context.File without validation. Attackers supplying a malicious kernel plugin can return an absolute path to read any file readable by the SiYuan process, including the API token and conf.json.

Craft CMS 4.0.0 before 4.18.7 and 5.0.0 before 5.10.14 passes the raw 'updates' request parameter in AppController::actionCacheUpdates to UpdatesModel without Component::cleanseConfig(), allowing behavior and event-handler injection. Authenticated control panel users with the performUpdates permission or Updates-utility access can supply 'as' or 'on' keys, such as an AttributeTypecastBehavior gadget, to execute arbitrary code.

Craft CMS 5.0.0 before 5.10.14 contains a DOM-based cross-site scripting vulnerability in the control panel login and 2FA verification flow, where a user-controlled returnUrl accepting javascript: URLs is used in client-side navigation. Attackers can lure a 2FA-enabled victim into opening a crafted /admin/login?verify=1 URL between password submission and TOTP verification, executing JavaScript in the authenticated control panel with the victim's permissions.

Craft CMS from 5.0.0 before 5.10.14 contains a missing authorization vulnerability in AppController::actionSwitchToLicensedEdition() that lets any authenticated user change the installed edition. Low-privileged front-end users can call the action route when the licensed edition mismatches to persistently downgrade the site edition in project config, which admins cannot reverse from the control panel.

Craft CMS from 5.10.2 before 5.10.14 contains a Twig sandbox bypass that allows non-admin users to invoke PHP callables through the "has some" and "has every" operators. Users with the System Messages permission can supply two-argument callables such as passthru or system in sandboxed templates to execute code as the web user.

Craft CMS 4.0.0 before 4.18.7 and 5.0.0 before 5.10.14 contains an arbitrary file read vulnerability because the Twig create() class blocklist does not cover DOMDocument. Attackers able to set an entry-type title format, including via project-config/apply, can use create('DOMDocument').loadHTMLFile() with textContent to expose local file contents to non-admin users who can view entries.

Parse Server before 8.6.93 and 9.0.0 before 9.10.1-alpha.13 does not enforce identity-scoped protectedFields (authenticated, role, or per-user groups) on count queries using the $relatedTo operator, exposing the size of protected Relation fields. An authenticated user who can read the owning object can combine counts with an objectId constraint as a membership oracle to reveal confidential memberships such as private groups or block lists.

Parse Server 8.2.2 before 8.6.94 and 9.0.0 before 9.10.1-alpha.14 applies its GraphQL introspection control only to operation text sent directly with a request, not to operations resolved from Apollo Server's automatic persisted query cache. Unauthenticated attackers holding the public Application Id and client key can replay an introspection query previously cached by a master or maintenance key client to retrieve the schema, including private class and field names.

Parse Server before 8.6.95 and 9.0.0 before 9.10.2-alpha.3 does not resolve file references lacking a URL before passing objects to Cloud Code triggers and LiveQuery, causing an uncaught exception that terminates the process. Attackers able to write to a class with Cloud Code save, find, or delete triggers or LiveQuery enabled, including unauthenticated users when public write access or sign-up is allowed, can crash the server or break subsequent requests.

Open WebUI from 0.6.31 before 0.11.4 contains a stored cross-site scripting vulnerability in its Markdown renderer, which treats user-authored tool-call details blocks as tool output and mounts their embeds attribute in script-enabled iframes. Authenticated users can post crafted channel messages, chat messages, or citation content that runs JavaScript in viewers' browsers, enabling in-app phishing, or session hijacking where iframeSandboxAllowSameOrigin is enabled.

Open WebUI from 0.9.0 before 0.11.4 contains a denial of service vulnerability that allows authenticated users with automations permission to stall the instance by submitting unsatisfiable recurrence rules. Attackers can repeatedly send rules like February 30th to the automations create or update endpoints, blocking the shared event loop for seconds per request.

Open WebUI 0.11.0 before 0.11.4 compiles user-chosen grep patterns for its knowledge and chat file search tools with a regex engine that expands nested counted repetition unbounded at compile time and caches the result. Authenticated users can have the model call grep_knowledge_files with short nested-repetition patterns, blocking the event loop and exhausting worker memory.