AgenticMail gives AI agents real email addresses and phone numbers. In @agenticmail/api prior to version 0.9.64, a low-privileged authenticated AgenticMail agent can enumerate another agent's pending/claimed tasks by supplying the target agent name to `GET /api/agenticmail/tasks/pending?assignee=<name>`. The returned task objects include the task IDs and payloads. The same task IDs can then be used with the capability-style task mutation endpoints (`/tasks/:id/claim`, `/tasks/:id/result`, `/tasks/:id/complete`, `/tasks/:id/fail`) to claim, complete, or fail tasks assigned to a different agent. Because ordinary authenticated agents can discover agent names through `GET /api/agenticmail/accounts/directory`, the task ID effectively stops being a secret capability. This turns the intended capability model into a cross-agent authorization bypass. Version 0.9.64 contains a fix.

FreeRDP before 3.28.0 treats lines beginning with forward slash in RDP files as raw command-line options, exposing the entire CLI parser surface to untrusted files. Attackers can craft malicious RDP files with /rdp2tcp, /cert:ignore, or /drive options to execute arbitrary commands, bypass certificate validation, or expose local filesystems without user interaction.

A vulnerability was identified in Metasoft 美特软件 MetaCRM up to 6.4.0 Beta06. The impacted element is an unknown function of the file /business/qnaire/upload.jsp. Such manipulation of the argument File leads to unrestricted upload. The attack may be launched remotely. The exploit is publicly available and might be used. The vendor was contacted early about this disclosure but did not respond in any way.

NextCRM is open-source customer relationship management (CRM) software. The CRM product catalog is an organization-wide business object. Normal application server actions restrict product creation, update, and deletion to `manager` and `admin` roles. However, in version 0.12.1, the MCP product tools expose the same write operations through `/api/mcp/mcp` using user-generated Bearer tokens and do not enforce role checks. Any authenticated low-privileged user who can generate an MCP API token can create, modify, archive, or soft-delete products in the shared CRM product catalog. Version 0.12.3 contains a fix.

NextCRM is open-source customer relationship management (CRM) software. In version 0.12.1, the MCP campaign tools expose campaign read and write operations over the network using user-generated Bearer API tokens (`nxtc__...`). The application has an authorization model that restricts normal users to campaigns they created, but multiple MCP campaign handlers ignore the authenticated user ID and query or mutate campaigns only by object ID. As a result, a low-privileged authenticated user with a valid MCP API token can enumerate all campaigns, read campaign details, update or delete campaigns owned by other users, modify campaign templates and steps, and potentially trigger or pause campaign delivery. Version 0.12.2 fixes the issue.

NextCRM is open-source customer relationship management (CRM) software. Versions prior to 0.12.0 have a Broken Object Level Authorization (BOLA/IDOR) vulnerability exists in the CRM contact and target update endpoints. The application fails to verify if the authenticated user has ownership of the specific resource being modified. This allows any authenticated user (even with a standard `member` role) to arbitrarily modify sensitive CRM contacts and targets belonging to other users or organizations (cross-tenant data tampering). Version 0.12.0 fixes the issue.

Path traversal on Windows in Apache MINA SSHD component sshd-git. Apache MINA SSHD is a Java library for client-side and server-side SSH. A git server implemented with Apache MINA SSHD component sshd-git and running on Windows could allow an authenticated remote user access to git repositories outside of the configured server-side root directory. The path validation applied for CVE-2026-48827 in Apache MINA SSHD 2.18.0 and 3.0.0-M4 was partly ineffective for Servers running on Windows. Applications are affected if they use org.apache.sshd:sshd-git to implement a git server and run on Windows. Applications not using sshd-git or not running on Windows are not affected. Users are advised to upgrade affected applications to Apache MINA SSHD 2.19.0, which fixes the issue. The issue also is present in the pre-release milestones 3.0.0-M1 to 3.0.0-M4 for a new upcoming new major version 3.0.0. Again, applications are affected only if they use sshd-git and run on Windows. Upgrade affected applications to 3.0.0-M5.

Improper certificate validation in Apache MINA SSHD (server-side). Apache MINA SSHD is a Java library for client-side and server-side SSH. Server-side OpenSSH user certificate validation during user authentication in an Apache MINA SSHD server did not check for the unsupported force-command or verify-required options that could be embedded in the certificate, nor did it validate these options. As a result it was possible that a user could authenticate with such a certificate that included a force-command option but still was able to execute other commands. What other command exactly would be available to the user depends on the implementation of the server. This issue is fixed in Apache MINA SSHD 2.19.0 and 3.0.0-M5. Applications are advised to upgrade to these versions. The fix rejects OpenSSH user certificates that include these options, since Apache MINA SSHD implements neither force-command nor sk-*-cert-v01@openssh.com user certificates (which are the only ones for which verify-required would make sense).

Path traversal in the sshd-scp component of Apache MINA SSHD. Apache MINA SSHD is a Java library for client-side and server-side SSH. The implementation of receiving files or directories via SCP did not validate filenames in SCP "C" or "D" commands. A malicious sender could send filenames containing paths, resulting in files to be written in attacker-controlled places. The issue affects only * applications that use no longer supported Apache MINA SSHD versions < 2.0.0 and use the SCP functions to receive files, * or applications using sshd-scp in Apache MINA SSHD >= 2.0.0 to receive files. Applications using Apache MINA SSHD >= 2.0.0 not using sshd-scp are not affected. The issue is fixed in Apache MINA 2.19.0 and 3.0.0-M5. Affected applications are advised to upgrade to these versions.

Huginn through 2022.08.18 contains a server-side request forgery vulnerability in the fetch_url method of ScenarioImport that allows authenticated users to make arbitrary HTTP requests by submitting crafted URLs. Attackers can probe internal network services, enumerate ports via error signatures, and access cloud metadata endpoints to retrieve sensitive credentials.

Glance through 0.8.5 contains an IP address spoofing vulnerability in the authentication handler that allows unauthenticated attackers to bypass brute-force lockout protections by supplying arbitrary values in the X-Forwarded-For request header when the server proxied option is enabled. Attackers can manipulate the leftmost value of the X-Forwarded-For header to make each login attempt appear to originate from a distinct IP address, preventing the per-IP failed-login counter from reaching the lockout threshold and enabling unlimited credential guessing against the authentication endpoint.

Adminer before 5.4.3 contains a cookie injection vulnerability that allows attackers to manipulate cookie attributes by injecting arbitrary values through the unsanitized X-Forwarded-Prefix HTTP header used in Set-Cookie path attributes. Attackers can exploit a misconfigured reverse proxy to downgrade SameSite protection and enable cross-origin authenticated requests, bypassing cookie security controls.

FileCodeBox before 2.4 contains a rate-limit bypass vulnerability in the IPRateLimit class that allows unauthenticated attackers to circumvent request throttling by supplying attacker-controlled X-Real-IP and X-Forwarded-For headers without verification of trusted reverse proxy origin. Attackers can supply unique spoofed IP values on each request to enumerate all possible share codes and retrieve other users' files without authentication.

HyperDX before 2.31.0 contains a server-side request forgery vulnerability that allows authenticated team members to direct the server to arbitrary internal destinations by supplying a caller-controlled host parameter to the ClickHouse proxy test endpoint with no URL validation or allowlist enforcement. Attackers can exploit the reflected error responses from the endpoint to disclose internal service response bodies, enabling access to internal APIs, container services, and cloud provider metadata endpoints.

Roo Code through 3.54.0 contains a command injection vulnerability in the auto-approve execute feature that allows attackers to bypass allowlist/denylist enforcement by nesting command substitutions inside parameter expansion defaults. The command parser in parse-command.ts replaces parameter expansions with opaque placeholders before extracting command substitutions, causing the containsDangerousSubstitution guard to miss nested payloads, which are then auto-approved based on the outer allowlisted command prefix and executed by the shell via execa, enabling arbitrary command execution.

DNG SDK versions 1.7.1 2536 and earlier are affected by a Stack-based Buffer Overflow vulnerability that could result in arbitrary code execution in the context of the current user. Exploitation of this issue requires user interaction in that a victim must open a malicious file.

LimeSurvey through 6.17.10 and 7.0.4 contains a server-side request forgery vulnerability in the REST API survey template endpoint that allows authenticated users to cause the server to issue arbitrary HTTP requests by supplying a manipulated Host header. Attackers can exploit the unsanitized use of the HTTP Host header in the getTemplateData() function to reach internal network services, cloud metadata endpoints, and extract sensitive credentials such as IAM tokens from instance metadata services.

A flaw was found in libcupsfilters and cups-filters. The PNG image reading function creates a libpng reader without installing an error recovery handler, causing the CUPS image filter process to abort when processing a malformed PNG file. An unauthenticated attacker could exploit this by submitting a specially crafted PNG print job, leading to denial of service of the in-flight print job.

FreeScout is a free help desk and shared inbox built with PHP's Laravel framework. Prior to version 1.8.221, FreeScout's attachment download route skips token authentication for any attachment whose `token_type` is set to `1` (`TOKEN_TYPE_LEGACY`). Because this route is unauthenticated and the file path is deterministic, an unauthenticated remote attacker can download any attachment that was created by an older version of FreeScout without possessing a valid token or session. Version 1.8.221 contains a fix.

xrdp is an open source RDP server. In versions 0.10.6 and prior, when an authenticated user session is initialized using the Xvnc backend over UNIX domain sockets, the Xvnc process is launched with insufficient authentication mechanisms. A local authenticated attacker could exploit this vulnerability to bypass intended session isolation, allowing them to unauthorizedly view or control the active desktop sessions of other users on the same system. Users using other backends, such as xorgxrdp or Xvnc over TCP sockets, are not affected. This issue has been fixed in version 0.10.6.1.

Chamilo version 1.11.40 and earlier are vulnerable to authenticated remote code execution in the main/inc/ajax/lang.ajax.php path. This endpoint is protected only by `api_protect_course_script(true)`, which means any authenticated user enrolled in a course (student, teacher, DRH) can reach it.

WhatsApp MCP Server is a Model Context Protocol (MCP) server for WhatsApp, enabling Claude to read and send WhatsApp messages. Prior to version 0.2.1, the `whatsapp-bridge` HTTP API listens on `127.0.0.1:8080` without authentication and without Host header validation, and the `/api/send` endpoint accepts an absolute `media_path` parameter without confining it to a safe directory. Combined, these issues allow any local process running as the same user as the bridge to send WhatsApp messages from the paired account without authorization; the same caller to read arbitrary files readable by the user (e.g. SSH private keys, browser session data, source code, dotfiles) and exfiltrate them as WhatsApp document attachments; and/or a remote attacker to trigger the same operations via DNS rebinding from a webpage the user visits, since no Host header validation is performed. In MCP environments, "local caller" extends beyond processes the user explicitly launched — sibling MCP servers, IDE extensions, and tool-triggered flows running in the user's session can act as the effective caller. This issue is fixed in whatsapp-mcp v0.2.1 and corresponding Docker images / release artifacts. Users should upgrade immediately. The fix introduces bearer token authentication on the bridge HTTP API (configured via environment variable, required on all requests, validated with constant-time comparison); host header allow-list validation to prevent DNS rebinding; and confinement of `media_path` to a configured directory, with rejection of absolute paths outside the root and path traversal sequences. This is a breaking change for clients of the bridge API. For users who cannot immediately upgrade: Stop the bridge, or block loopback access to port 8080, when the bridge is not actively in use; avoid running the bridge alongside untrusted MCP servers, browser extensions, or other untrusted local processes; avoid browsing untrusted sites while the bridge is running (DNS rebinding mitigation); and/or run the bridge under a dedicated user account or in a sandbox/container with no access to sensitive files.

Network-AI is a TypeScript/Node.js multi-agent orchestrator. Prior to version 5.12.2, `EnvironmentManager.listBackups()` reads each backup's `_manifest.json` and trusts the manifest's `path` field. `EnvironmentManager.pruneBackups()` later passes that trusted `entry.path` directly to `rmSync(entry.path, { recursive: true, force: true })`. An attacker who can place or modify a manifest inside `data/<env>/.backups/<name>/_manifest.json` can cause `network-ai env backup prune --env <env> --keep <n>` or any code path invoking `pruneBackups()` to recursively delete an arbitrary path accessible to the Network-AI process user. This is fixed in v5.12.2. `pruneBackups()` no longer passes `entry.path` from the on-disk manifest to `rmSync`. The deletion path is recomputed from a format-validated `entry.backupId`, and a `dirname` containment check confines deletion to exactly one level under the backups directory. A poisoned manifest (e.g. `"path": "/"`) is now inert.

xrdp is an open source RDP server. In versions 0.10.6 and prior, a n issue was discovered where the software fails to properly validate the totalLength field within the RDP protocol control header during packet reception. An unauthenticated remote attacker can exploit this vulnerability by sending a specially crafted packet that forces the xrdp process or thread into an infinite, CPU-bound loop. Because the internal pointer fails to advance and the deadlock prevention mechanism is bypassed for specific protocol data unit types, the process consumes excessive CPU resources indefinitely. This can render the xrdp service unavailable and potentially lead to system-wide resource exhaustion if multiple malicious connections are established. This issue has been fixed in version 0.10.6.1.

Due to a missing sanitization call in [`afsql_dd_run_query`](https://github.com/syslog-ng/syslog-ng/blob/649e6e18e3459fb4467000a88dfb12fa97f9719c/modules/afsql/afsql.c#L219), syslog-ng before 4.12 are vulnerable to SQL injection from an untrusted source. This is not part of the default configuration, the SQL driver has to be manually configured. Fixes are in syslog-ng 4.12, syslog-ng Premium Edition 8.2 and syslog-ng Store Box 7.8

Frappe LMS is an open source learning management system. In version 2.51.0 and earlier, a user could bypass payment validation for courses by using unrelated batch. This has been patched in 2.52.0 with enrollment now validating that the batch is linked to course.

Network-AI is a TypeScript/Node.js multi-agent orchestrator. Prior to version 5.4.5, the MCP SSE server defaults to an empty secret (`process.env['NETWORK_AI_MCP_SECRET'] ?? ''` at `bin/mcp-server.ts:89`), which causes `_isAuthorized` (`lib/mcp-transport-sse.ts:254`) to return `true` unconditionally for every request — no `Authorization` header is required. Simultaneously, `_handleRequest` sets `Access-Control-Allow-Origin: *` (`lib/mcp-transport-sse.ts:272`) on every response, so a cross-origin browser fetch can read the result without restriction. An unauthenticated attacker who can lure a user to a malicious web page can invoke all 22 exposed MCP tools — including `config_set`, `agent_spawn`, and `blackboard_write` — against a default-configured localhost server. Version 5.4.5 patches the issue.

libvips is a fast image processing library with low memory needs. The `tiffload` operation in libvips versions before and including 8.18.1 could incorrectly determine the number of channels in a JPEG or JPEG2000-encoded tile within a TIFF image, leading to a possible buffer overflow. This has been patched in version 8.18.2.

libvips is a fast image processing library with low memory needs. The `vipsload` operation in versions before and including 8.18.0 could incorrectly determine image dimensions leading to an integer overflow and a subsequent heap-based buffer overflow. This has been patched in version 8.18.1.

dataCycle is a data management system for centrally storing, managing, searching, finding, and distributing data. In dataCycle-CORE, the module handling core processing and framework rules, before and including version 25.07.3, the application accepts unlimited password guesses against both the browser login flow and the JSON login endpoint. The source code enables Devise's `:lockable` module on the user model but explicitly disables both lock and unlock strategies, and no request throttling or rate-limiting layer was identified in the Rails code. This creates a direct online password-guessing risk: - valid user accounts can be attacked continuously without temporary lockout - the same weakness is reachable through both `/users/sign_in` and `/api/v4/auth/login` - successful guessing yields a normal session cookie in the HTML flow or a fresh JWT in the API flow - the API endpoint is especially attractive for automation because it requires no CSRF token This has been patched in version 26.06.08.

dataCycle is a data management system for centrally storing, managing, searching, finding, and distributing data. In dataCycle-CORE, the module handling core processing and framework rules, before and including version 25.07.3, a low-privileged authenticated API user can supply `forwardToUrl` and `redirectUrl` values when triggering password reset or confirmation flows. Those values are then embedded into the outgoing email workflow without host allowlisting. This creates two related abuse paths: - password reset or confirmation links can be sent to a victim with the token already attached to an attacker-controlled `forwardToUrl` - after a legitimate password reset completes, the browser is redirected to attacker-controlled `redirectUrl` In practice, this can be used for phishing, token capture, confirmation hijacking, or steering a victim from a trusted email into an attacker domain. This is patched in version 26.06.08.

dataCycle is a data management system for centrally storing, managing, searching, finding, and distributing data. In dataCycle-CORE, the module handling core processing and framework rules, before and including version 25.07.3, any authenticated user can request arbitrary partials or helper-backed render functions through /remote_render. The endpoint does not restrict which partial can be rendered and does not apply controller-specific authorization before rendering the selected view. This enables a low-privileged user to retrieve server-side rendered admin content that is otherwise hidden by navigation and route checks. On the test instance, a Standard user was able to retrieve the PostgreSQL admin dashboard stats even though /admin itself redirected away. This is patched in 26.06.08.

dataCycle is a data management system for centrally storing, managing, searching, finding, and distributing data. In dataCycle-CORE, the module handling core processing and framework rules, before and including version 25.07.3, the documentation and static markdown renderer accepts attacker-controlled path segments and only runs them through the Rails HTML sanitizer, which does not remove directory traversal sequences. An unauthenticated attacker can traverse out of the intended `docs` or `static` directories and render arbitrary `.md` files from the application root or engine root. This is patched in version 26.06.08.

dataCycle is a data management system for centrally storing, managing, searching, finding, and distributing data. In dataCycle-CORE, the module handling core processing and framework rules, before and including version 25.07.3, anyone with a DataLink UUID can fetch the attached text file directly, even if the link is expired, the caller is unauthenticated, or the normal show flow would have denied access. Because the route is public and the mailer embeds the direct file URL, any leaked, forwarded, logged, or stale email link can continue to expose the attachment.

Mailpit is an email testing tool and API for developers. Prior to version 1.30.0, the Mailpit SMTP server has a Server.MaxSize int field that controls the maximum allowed DATA payload size, but the field is never assigned anywhere outside test code, leaving it at Go's zero value (0 ⇒ "no limit"). The same applies to the HTTP /api/v1/send endpoint, whose request body is decoded with json.NewDecoder(r.Body) and no http.MaxBytesReader. Because Mailpit's default listeners bind [::]:1025 (SMTP) and [::]:8025 (HTTP), with no authentication required on either, a single network-reachable attacker can push an arbitrarily large message into Mailpit and watch RAM consumption spike with a ~7-10× amplification factor (raw frame → enmime envelope tree → search-text index → zstd-encoded write to SQLite). Repeating the attack — or running it concurrently from multiple connections — drives the process to OOM-kill. Version 1.30.0 contains a patch.

ProFTPD before 1.3.9c and 1.3.10rc3 contains a signed integer overflow vulnerability in the mod_sftp module's SCP size-record parser that allows authenticated low-privilege attackers to bypass ASLR by sending a crafted file size value of UINT64_MAX, which results in a negative off_t value. Attackers can exploit the subsequent conversion to uint32_t, causing an approximately 4 GB requested read length and forcing the server to read beyond the end of the SSH channel data and write overread process memory into the uploaded file. In tested configurations, the disclosed data contains libc, libcrypto, and PIE pointers sufficient to derive their randomized base addresses, thereby bypassing ASLR and enabling reliable exploitation of memory corruption vulnerabilities in the same process.

FileBrowser Quantum is a free, self-hosted, web-based file manager. Prior to version 1.4.3-beta, the `subtitlesHandler` endpoint (`GET /api/media/subtitles`) accepts two user-controlled query parameters: `path` and `name`, both of which are used in filesystem operations without sanitization, creating two independent path traversal vectors. The primary vector is the `path` parameter: it is passed directly to `idx.GetRealPath()` without calling `SanitizeUserPath()`, allowing an attacker to escape the storage root and set `parentDir` to any directory on the host. No existing anchor file is required. The secondary vector is the `name` parameter: it is joined with `parentDir` via `filepath.Join(parentDir, name)` without stripping directory components, allowing traversal relative to any resolved `parentDir`. Any authenticated user (regardless of role or permissions) can exploit either vector to read any text file readable by the server process, including `/etc/passwd`, SSH keys, database credentials, and JWT signing keys. Version 1.4.3-beta patches the issue.

A security flaw has been discovered in Beijing Shenzhou Shihan Technology Multimedia Integrated Business Display System 8.2.2. Impacted is an unknown function of the file /admin/system/structure/updateStructure/deflate/Insecure/Staffshinel Ds.jsp?Shine ID=aaa. The manipulation of the argument Structure_ID results in sql injection. It is possible to launch the attack remotely. The exploit has been released to the public and may be used for attacks.

A vulnerability was found in Tenda AC10 16.03.10.09_multi_TDE01. This issue affects the function fromAdvSetLanip of the file /goform/AdvSetLanip of the component httpd/netctrl. The manipulation of the argument GetValue/SetValue results in stack-based buffer overflow. The attack may be performed from remote. The exploit has been made public and could be used.

A flaw was found in the QEMU Guest Agent (qga). A local unprivileged user can exploit a vulnerability in the guest-ssh-add-authorized-keys command handler by manipulating symbolic links. This can occur either through a deterministic directory-symlink bypass or a Time-of-Check to Time-of-Use (TOCTOU) file-symlink race. Successful exploitation allows the attacker to gain ownership of arbitrary root-owned files or directories, leading to root access. This vulnerability requires an external management layer (e.g., libvirt) to trigger the affected code path.

Network-AI (npm: network-ai) versions 5.12.2 through 5.13.3 fail to apply the configured authorization check (checkAuth/secret) to the ApprovalInbox GET read routes, so even when an operator configures a secret, unauthenticated actors can access sensitive approval request details. The GET /approvals/?status=all, GET /approvals/:id, GET /approvals/stats, and GET /approvals/sse routes disclose full ApprovalEntry content including action/target shell-command strings, file paths, justifications, and risk levels. All responses also carry a hardcoded Access-Control-Allow-Origin: * header, enabling cross-origin disclosure from any website the operator visits. This is an incomplete fix for GHSA-mxjx-28vx-xjjj.

FreeRDP before 3.28.0 (affected 3.x through 3.27.1) contains a double-free vulnerability in freerdp_client_rdp_file_apply_to_settings() (client/common/file.c) when parsing the selectedmonitors field of a .rdp connection file. The MonitorIds array is allocated through the settings object, and a raw non-owning pointer to it is freed on the strtoul error path without clearing settings->MonitorIds, leaving it dangling; at teardown freerdp_settings_free() frees the same buffer again. An attacker who convinces a victim to open a crafted .rdp file with oversized monitor tokens can trigger a size-controlled double-free in any FreeRDP CLI client (xfreerdp/sdl-freerdp/wlfreerdp) in the default configuration.

SurrealDB before 2.5.0 and before 3.0.0-beta.3 contains a confused deputy privilege escalation vulnerability. Unprivileged users (e.g., those with the database editor role) can create or modify fields containing futures, functions, or closures. Because these are executed in the context of the invoking/querying user rather than their creator, an attacker can plant malicious logic that executes with a higher-privileged user's permissions when that user reads or writes the affected record. This can lead to full privilege escalation, including creation of a root owner and server takeover.

SurrealDB before 3.1.0 fails to enforce the configured recursion depth limit in the value and JSON parser when processing nested braces, brackets, or parentheses. Unauthenticated attackers can send deeply nested JSON payloads to the WebSocket /rpc endpoint to exhaust server memory and crash the process.

SurrealDB before 3.1.0 fails to enforce recursion depth limits in the type/kind parser when processing nested type annotations. Authenticated attackers can send queries with deeply nested type annotations to exhaust server memory and crash the process.

SurrealDB before 3.1.0 evaluates user-supplied WHERE clauses in SELECT statements (and SET/MERGE/CONTENT/PATCH clauses in UPDATE, UPSERT, INSERT ON DUPLICATE KEY UPDATE, and RELATE update-variant statements) against full record data before enforcing PERMISSIONS FOR SELECT WHERE restrictions. An authenticated user — including Record and Scope users — can exploit this ordering flaw to read the full contents of any table in the database they are authenticated against, bypassing table-level permission checks. Exfiltration is most direct when scripting functions are enabled (--allow-scripting), but is also possible via SurrealQL's THROW statement and timing-based side channels without scripting. The vulnerability is confined to the attacker's current database and does not cross namespace or database isolation boundaries.

SurrealDB versions before 3.1.0 contain a denial of service vulnerability where malicious LIVE queries with WHERE clauses that evaluate to errors cause all CREATE, UPDATE, and DELETE operations on the watched table to fail. An authenticated user with only select permission can prevent write operations on a table for any user, including root, by registering a LIVE query that triggers evaluation errors until the query is killed or the session ends.

SurrealDB versions before 3.1.0 contain a denial of service vulnerability in the RPC use handler that panics when db is set without a namespace. Unauthenticated attackers can send a malformed WebSocket message to the /rpc endpoint to crash the server process.

SurrealDB versions before 3.1.0 fail to enforce table SELECT permissions when traversing graph edges or back-references. Authenticated users can read records from any table reachable through graph edges regardless of the target table's PERMISSIONS FOR select clause.

SurrealDB versions before 3.1.4 fail to properly enforce SELECT permissions on array elements (field.*) for record users, leaking denied array elements instead of hiding them. Attackers with record scope access can read array elements that element-level permissions should deny by exploiting incorrect index handling during permission filtering.