By severity

High-severity vulnerabilities

CVEs rated High by CVSS, with SEC.co remediation and prioritization guidance.

4140 published vulnerabilities · page 17 of 42

• CVE-2026-48725HIGH 8.1

  Warp is a terminal development environment that allows developers to work more efficiently with code and terminal commands. A vulnerability in Warp versions from April 2021 through early May 2026 allows malicious terminal output—such as from a compromised remote server or script—to silently read or write data to your system clipboard without asking for permission first. An attacker could craft malicious terminal output that steals sensitive information you've copied (like API keys or passwords) or inject malicious content into your clipboard for you to paste elsewhere. The vulnerability was patched in version 0.2026.05.06.15.42.stable_01.

• CVE-2026-49121HIGH 8.1

  AI Tensor Engine for ROCm (AITER) versions up to 0.1.14 contain a critical flaw that allows attackers to run arbitrary code on inference worker machines without authentication. An attacker who can reach certain network endpoints or forge specific credentials can send a specially crafted message that executes code with the privileges of the inference worker process across multiple nodes simultaneously. This vulnerability requires network access to specific cluster communication channels but bypasses all authentication and validation mechanisms.

• CVE-2026-49286HIGH 8.1

  PhpWeasyPrint, a PHP library for converting URLs and HTML pages into PDF files, contains a critical flaw that allows attackers to execute arbitrary code on servers running vulnerable versions. The library was supposed to block a dangerous file access method called `phar://`, but the protection was case-sensitive, meaning attackers could bypass it by using variations like `PHAR://` or `Phar://`. When an attacker tricks the application into processing a specially crafted PHAR archive, it triggers automatic code execution during deserialization on PHP 7 versions. This is a repeat of an earlier vulnerability that should have been fixed; the library was not comprehensive in its original patch. Upgrading to version 2.6.0 or later closes this hole.

• CVE-2026-49291HIGH 8.1

  mcp-memory-service, a semantic memory layer used in AI applications, contains an authorization flaw that allows attackers with read-only access to perform write operations. The vulnerability exists in the HTTP MCP JSON-RPC endpoint at `/mcp`, which fails to enforce proper permission checks when dispatching tool calls. Specifically, authenticated users granted only read-level OAuth scope can invoke the `store_memory` and `delete_memory` functions despite those operations being restricted to write-scope users at the REST API layer. This permission bypass enables unauthorized data modification in memory stores that should be protected.

• CVE-2026-49297HIGH 8.1

  Apache Airflow's Google Cloud Storage integrations contain a path traversal vulnerability that allows malicious object naming to write files outside their intended destination. If your DAGs ingest data from GCS buckets controlled by external partners, service accounts, or public sources, an attacker could craft object names with `..` sequences to overwrite files on SFTP servers or worker machines. The risk is elevated in multi-tenant or federated data environments where bucket write permissions are granted to less-trusted principals.

• CVE-2026-49340HIGH 8.1

  Gonic, a free-software music streaming server that implements the Subsonic API, contains a vulnerability in how it handles playlist creation and updates. Any authenticated user—not just administrators—can exploit a flaw in the playlist write function to place files anywhere on the server's filesystem and create directories with overly permissive access controls. This was patched in version 0.21.0.

• CVE-2026-49402HIGH 8.1

  Deno's implementation of Node.js child process spawning on Windows contains a command injection vulnerability in its shell argument escaping logic. When developers use the shell:true option with spawn, spawnSync, exec, or related functions, Deno fails to properly quote and escape arguments containing Windows cmd.exe metacharacters, particularly the percent sign (%). An attacker who can control any part of an argument passed to these functions can inject and execute arbitrary additional commands within the spawned cmd.exe process. This affects Deno versions before 2.7.10 and is resolved in that release.

• CVE-2026-49490HIGH 8.1

  OpenCATS, a recruitment applicant tracking system, contains an SQL injection flaw affecting versions from 0.9.1a onward. An attacker with valid login credentials can craft malicious filter requests in the Candidates DataGrid to bypass security controls and inject SQL commands into the database. The vulnerability specifically exploits the Tags column, which is marked as non-filterable but can be targeted through specially crafted requests. Once injected, attackers can read, modify, or delete sensitive candidate and company data stored in the database.

• CVE-2026-49872HIGH 8.1

  Apache APISIX versions 3.0.0 through 3.16.0 contain an authentication bypass flaw in the CAS authentication plugin. An attacker with valid credentials to one system can potentially use those credentials to authenticate as a different user or gain unauthorized access through the misconfigured authentication logic. This is a significant issue because CAS (Central Authentication Service) is commonly deployed in enterprise environments where authentication decisions directly gate access to sensitive APIs and services.

• CVE-2026-49877HIGH 8.1

  Apache ActiveMQ contains an improper authorization flaw that allows any authenticated user with low-privilege access to the Web Console to reach administrative paths they shouldn't be able to access. By default, the embedded Jetty web server fails to enforce proper access controls on /admin/* endpoints, meaning a regular user can potentially perform administrative actions. This affects ActiveMQ versions before 5.19.8 and 6.0.0 through 6.2.6.

• CVE-2026-49948HIGH 8.1

  Mem0 through version 0.2.8 has a privilege escalation vulnerability in its self-hosted server. Any user with an API key can reconfigure the global AI model and embedding providers to point to an attacker-controlled server. The attacker doesn't need elevated permissions—just normal API access. The malicious configuration persists in the database and survives restarts, affecting all users on that Mem0 instance. This allows the attacker to intercept, log, or manipulate all LLM requests and responses.

• CVE-2026-50101HIGH 8.1

  Naxclow relay devices contain a critical credential management flaw that compromises long-term security. Each device is issued a server-side relay credential at boot time that never changes and cannot be revoked—even by the device owner. If an attacker gains access to this credential through any means (network interception, leaked documentation, compromised backups, or previous breaches), they retain the ability to impersonate the device indefinitely. Factory resets and device re-onboarding do not invalidate the compromised credential, meaning a single credential exposure can lead to persistent unauthorized access to the device's relay channel.

• CVE-2026-50107HIGH 8.1

  NGINX Gateway Fabric contains a configuration injection vulnerability in how it processes user-supplied access log format settings. When an authenticated user modifies the NginxProxy resource definition, their input is directly inserted into NGINX configuration files without proper validation. An attacker with control-plane access can inject malicious NGINX directives, potentially compromising system integrity and confidentiality. This is a control-plane issue, not a data-plane vulnerability, but it creates a significant security boundary violation for multi-tenant or role-based access control scenarios.

• CVE-2026-50632HIGH 8.1

  Apache CXF contains a gap in a previously patched vulnerability that allows remote code execution when untrusted users can configure Java Message Service (JMS) endpoints. This is a second-order fix to CVE-2026-44417, indicating the original patch was incomplete. An attacker with JMS configuration privileges could execute arbitrary code on the system, making this a serious risk for organizations relying on Apache CXF to handle messaging.

• CVE-2026-50633HIGH 8.1

  Apache CXF, a widely-used web services framework, contains a JNDI (Java Naming and Directory Interface) injection flaw in its JCA integration module. An attacker who can modify the JCA deployment descriptor file (ra.xml) or influence runtime activation parameters could inject malicious code that executes with the privileges of the application server. This is a high-risk vulnerability because JNDI injection can lead to full code execution and system compromise.

• CVE-2026-50721HIGH 8.1

  Libreswan, a widely-used open-source IPsec implementation, contains a cryptographic signature verification flaw in its IKEv1 packet handling. When processing authentication signatures (SIG payloads), the library fails to properly validate the length of cryptographic hashes, creating two distinct attack vectors: an attacker with knowledge of small RSA public exponents (such as e=3) can forge authentication credentials and impersonate legitimate peers, while a separate attack path allows remote denial of service by sending malformed signatures that trigger an assertion failure. When exploited for denial of service, the Libreswan daemon crashes and automatically restarts, but repeated exploitation causes sustained outages. The vulnerability does not enable remote code execution, and X.509 certificate-based peer verification remains unaffected.

• CVE-2026-50722HIGH 8.1

  Libreswan, a widely-used open-source IKE/IPsec implementation, contains a flaw in how it validates digital signatures on IKEv2 authentication payloads. The vulnerability stems from improper verification of the ASN.1 encoding structure within RSA signatures. An attacker can exploit this in two ways: first, by forging authentication messages when the target uses a small RSA public exponent (such as e=3), allowing them to impersonate legitimate IKE peers; second, by sending specially crafted malformed signatures that crash the daemon, causing repeated denial of service. While serious, this does not enable remote code execution, and X.509 certificate chains remain properly validated.

• CVE-2026-50874HIGH 8.1

  kanishka-linux Reminiscence version 0.3.0 contains a command injection flaw in its media management feature that allows authenticated users to execute arbitrary operating system commands on the affected server. An attacker with valid credentials can craft malicious input to the /manage/features/media endpoint to run commands with the privileges of the web application process, potentially leading to full system compromise.

• CVE-2026-50875HIGH 8.1

  Deck9 Input v2.0.1 contains a flaw in how it controls access to webhook management endpoints. An attacker with valid credentials to one tenant account can craft requests to modify or delete webhooks belonging to a different tenant. This cross-tenant privilege escalation allows unauthorized tampering with webhook configurations—potentially redirecting data flows, disabling integrations, or causing service disruption—without requiring administrative access or special exploitation techniques.

• CVE-2026-50881HIGH 8.1

  Bonsai v6.0 contains a privilege escalation vulnerability that allows authenticated users with Editor-level permissions to gain full Administrator access. Once escalated, attackers can make unauthorized changes to accounts, passwords, and system configurations. The vulnerability requires an attacker to already have valid Editor credentials, but no additional user interaction is needed to exploit it.

• CVE-2026-50888HIGH 8.1

  Benjamin Jonard Koollection version 1.8.0 contains a security flaw in its scraper feature that allows authenticated users to trick the application into accessing internal resources they shouldn't reach. An attacker with valid login credentials can supply specially crafted URLs to the scraper, forcing the server to make requests to internal systems—such as private databases, cloud metadata services, or other backend infrastructure—and potentially retrieve sensitive data. This vulnerability requires authentication, so it's most dangerous in environments where user accounts are compromised or where internal users pose a threat.

• CVE-2026-50891HIGH 8.1

  Filestash v0.4.0 contains an access control vulnerability in its admin configuration API endpoint that allows authenticated users to escalate their privileges beyond their intended permission level. An attacker with valid login credentials can craft a specially formed request to /admin/api/config to gain elevated privileges, potentially accessing or modifying sensitive system settings reserved for administrators.

• CVE-2026-5120HIGH 8.1

  BIOVIA Workbook contains a race condition—a timing-based flaw where two operations can interfere with each other—that allows an authenticated user to view data belonging to another user. The vulnerability affects all versions from Release 2021 through Release 2026. Because an attacker needs valid login credentials, the risk is primarily from insiders or compromised accounts within your organization. The flaw does not allow data modification or system disruption, only unauthorized viewing.

• CVE-2026-51923HIGH 8.1

  docuForm GmbH Client version 11.11c contains a flaw that allows authenticated users to access and control other users' accounts without authorization. An attacker with valid login credentials can manipulate the user settings feature to execute code, steal sensitive information from other accounts, or alter data they should not have access to. The vulnerability requires an attacker to already be logged in, but once inside, the lack of proper access controls makes it easy to escalate privileges and compromise multiple user accounts.

• CVE-2026-51924HIGH 8.1

  A vulnerability in docuForm GmbH Client version 11.11c allows attackers with user-level access to upload files and execute arbitrary code on affected systems. The attack requires authentication but no additional interaction from the target, making it a reliable path to full system compromise once an attacker has gained initial credentials.

• CVE-2026-51925HIGH 8.1

  A vulnerability in docuForm GmbH Client version 11.11c allows authenticated attackers to read arbitrary files from the server and potentially execute code. The flaw exists in the dfm-menu_report.php component and can be exploited remotely by an attacker with valid login credentials. An attacker could use this to extract sensitive data such as database credentials, configuration settings, or source code—information that could enable further attacks or data theft.

• CVE-2026-52801HIGH 8.1

  Gogs, a self-hosted Git service, contains a vulnerability in its Mirror Settings feature that allows authenticated users to bypass intended security controls and import local repositories without proper validation. The flaw exists because the SaveAddress function fails to adequately validate user input, creating an alternative attack path that circumvents the more secure New Migration functionality. This issue affects Gogs versions prior to 0.14.3 and requires authentication to exploit.

• CVE-2026-52845HIGH 8.1

  Caddy, a modern web server platform, contains a header-handling vulnerability that allows authenticated attackers to inject fake identity or group information into requests sent to PHP applications. The issue stems from a mismatch in how Caddy processes headers during authentication checks versus how it later converts those headers for PHP/FastCGI backends. An attacker can exploit this by sending a specially crafted header variant that bypasses Caddy's security controls, potentially allowing privilege escalation or impersonation within downstream applications. The flaw affects versions before 2.11.4 and is corrected in that release.

• CVE-2026-52967HIGH 8.1

  A memory safety flaw in the Linux kernel's SMB client code can cause the system to hang indefinitely or read sensitive data from memory when processing specially crafted symbolic link responses. The vulnerability affects 32-bit systems more directly, but the underlying logic issue exists in the code path. An attacker would need to trick a user into connecting to a malicious SMB server or intercept and modify SMB traffic to exploit this.

• CVE-2026-53147HIGH 8.1

  A flaw in the Linux kernel's Thunderbolt XDomain protocol handler allows an attacker on the same network segment to crash a system or potentially read sensitive memory. When a Thunderbolt peer device sends a specially crafted packet, the kernel accepts it based on a generic header size check but then attempts to interpret the data as larger protocol-specific structures without verifying the packet is actually long enough. This mismatch causes the kernel to read beyond the packet boundary, potentially exposing kernel memory contents or triggering a system halt.

• CVE-2026-53178HIGH 8.1

  A memory safety bug in the Linux kernel's WiFi driver for certain Realtek devices could allow an attacker on the local network to crash the system or potentially leak sensitive information. The issue stems from missing validation that could lead to integer underflow—a condition where subtracting a number from an unsigned integer causes it to wrap around to an extremely large value, corrupting memory operations. Devices must be on the same network segment as the attacker, but no user authentication or interaction is required for exploitation.

• CVE-2026-53254HIGH 8.1

  A vulnerability in the Linux kernel's Bluetooth RFCOMM (Radio Frequency Communication) protocol handler allows a nearby attacker to read sensitive memory from a vulnerable device by sending specially crafted Bluetooth frames. The kernel fails to check the size of incoming data before processing it, leading to out-of-bounds memory reads. An attacker within Bluetooth range can exploit this without authentication to potentially leak confidential information or crash the system.

• CVE-2026-53407HIGH 8.1

  Zoom Workplace for mobile devices contains an authorization flaw in how it handles custom URL schemes. An attacker with network access and valid Zoom credentials could craft a malicious link or request that bypasses intended security controls, allowing them to escalate their privileges within the application. The vulnerability affects Android versions before 7.0.4 and iOS versions before 7.0.3. While this requires some form of authentication to exploit, the impact allows an attacker to gain elevated access they shouldn't normally have.

• CVE-2026-53408HIGH 8.1

  Zoom Workplace for Android and iOS contains a flaw in how it handles custom URL schemes, allowing someone with network access to bypass normal authorization checks and gain elevated privileges on the device. An unauthenticated attacker can craft a malicious link or redirect that exploits this improper authorization logic, potentially gaining access to sensitive features or data within the app without proper authentication. This affects Android versions before 7.0.4 and iOS versions before 7.0.3.

• CVE-2026-53673HIGH 8.1

  BuddyPress 14.4.0 has a critical flaw in how it handles private message requests through its REST API. An authenticated attacker can manipulate the request to view, reply to, or delete private messages belonging to any other user on the platform. The vulnerability stems from inadequate validation—the system checks whether the user ID provided in the request has access, rather than verifying the logged-in user should have that access. This means an attacker with any valid account can impersonate others and gain full control over their private message threads.

• CVE-2026-53738HIGH 8.1

  Copy & Delete Posts, a WordPress plugin, contains a critical authorization flaw that allows any user with a plugin-enabled role—even non-administrators—to perform privileged operations they shouldn't be able to access. Through a parameter called 'f', attackers can delete posts or change plugin configuration settings. The vulnerability stems from missing per-function permission checks in the plugin's AJAX handler, meaning the plugin trusts user input without verifying what actions each user is authorized to perform.

• CVE-2026-53777HIGH 8.1

  Perry, a build or artifact management tool, has a path traversal vulnerability affecting versions before 0.5.1159. When a malicious build server sends specially crafted messages to a Perry client, it can trick the client into writing files to unintended locations on the system. This could allow an attacker controlling the server to overwrite important files or steal sensitive data from the client machine. The vulnerability requires user interaction—the client must be actively connected to and communicating with the compromised server—but once exploited, the impact is significant.

• CVE-2026-53823HIGH 8.1

  OpenClaw versions prior to 2026.5.3 contain a privilege escalation flaw in how the allowFrom access control feature validates user identity. The vulnerability stems from binding authorization decisions to Slack display names, which are user-mutable metadata. An attacker with any valid Slack account can change their display name to impersonate a different user's identity policy entry, bypassing intended access restrictions and gaining unauthorized agent access. This is particularly dangerous because display name changes require no special privilege and can be executed repeatedly to probe for matching policy entries.

• CVE-2026-53849HIGH 8.1

  OpenClaw versions before 2026.5.7 contain a privilege escalation flaw in the allowFrom access control feature. The vulnerability stems from relying on Discord display names—which users can freely change—to authorize agent access instead of using immutable Discord user IDs. An attacker can simply rename their Discord account to match an authorized identity in the policy, bypassing intended access restrictions and gaining unauthorized control over agents meant for another user.

• CVE-2026-53855HIGH 8.1

  OpenClaw versions before 2026.4.2 contain a security flaw that allows authenticated operators to bypass security controls meant to restrict which commands can be executed. By exploiting how shell parameters work, an attacker with valid credentials can combine approved tools with special shell arguments to sneak in and run commands that should be blocked. This creates a significant risk because it defeats the intended protection layer.

• CVE-2026-53857HIGH 8.1

  OpenClaw versions prior to 2026.5.3 have a flaw in how they enforce access control policies for Zalo contacts. Attackers who can change their display name in Zalo can trick the system into sending them agent responses that were meant for other users. This happens because the policy engine relies on display name metadata to decide who should receive sensitive information, but attackers can modify their display name after initially connecting. An authenticated attacker could exploit this to receive privileged communications intended for different Zalo identities.

• CVE-2026-53864HIGH 8.1

  OpenClaw versions before 2026.5.26 have a security weakness in how they clean up environment variables before running processes. An attacker who can modify workspace configuration files (.env files), override tool settings, or edit skill blocks can inject specially crafted Node.js control variables that will be passed through to child processes or manipulate where code coverage reports are saved. This could allow unauthorized access to sensitive data or modification of application behavior.

• CVE-2026-53866HIGH 8.1

  OpenClaw, a tool used by authenticated operators, has a flaw in how it processes shell commands. An attacker with valid credentials could bypass the system's approval controls and execute commands that should have been blocked. The issue stems from a parsing gap where certain command formats skip the required allowlist check, letting unauthorized shell commands run without triggering approval prompts.

• CVE-2026-53871HIGH 8.1

  Hermes WebUI versions before 0.51.368 have an authorization flaw that allows an authenticated attacker to impersonate other user profiles. The vulnerability exists in how the application validates the hermes_profile cookie—it trusts user-supplied profile names without properly verifying that the requester owns or has permission to access those profiles. An attacker who is already logged in could modify this cookie to view, modify, or access files and sessions belonging to other users on the system.

• CVE-2026-53903HIGH 8.1

  MCO (MyComplianceOffice) contains a critical authorization flaw that allows authenticated users to view trading documents belonging to other customers. An attacker with valid login credentials can request documents from different users by manipulating document identifiers in the web API endpoint, potentially exposing sensitive financial and compliance records. While the attack requires knowing or guessing valid document IDs, the predictable nature of these identifiers makes systematic enumeration feasible.

• CVE-2026-5422HIGH 8.1

  A flaw in Jupyter Server version 2.17.0 allows authenticated users to read and write files outside their intended directory boundaries. The vulnerability stems from an incomplete security check that fails to properly validate file path requests, allowing attackers to access sibling directories and potentially sensitive files in shared hosting environments. An attacker with login credentials can exploit this to breach confidentiality and integrity of data across directory structures that were meant to be isolated.

• CVE-2026-54415HIGH 8.1

  Azuriom CMS versions before 1.2.11 contain a missing authorization flaw in server management routes that allows authenticated administrators with the admin.access permission to escalate privileges and take over regular user accounts. An attacker with admin credentials can create AzLink server tokens and use API endpoints to forcibly change user passwords and email addresses, effectively locking legitimate users out of their accounts and taking complete control of them.

• CVE-2026-54512HIGH 8.1

  jackson-databind, a widely-used Java library for converting between JSON and Java objects, contains a critical bypass of its built-in polymorphic type validation. When an application deserializes JSON with polymorphic types enabled, attackers can craft a type identifier that includes generic parameters (e.g., ArrayList<EvilClass>) to slip denied classes past the safety check. The validator only checks the outer container type (ArrayList) but ignores what's inside the angle brackets, allowing instantiation and exploitation of restricted classes. This affects versions from 2.10.0 through 2.18.7, 2.21.3, and 3.1.3.

• CVE-2026-54513HIGH 8.1

  Jackson-databind's polymorphic type validation (PTV) can be bypassed when using the `allowIfSubTypeIsArray()` method combined with a concrete-type allowlist. The validator fails to check the element type of arrays, allowing instantiation of non-allowlisted classes wrapped in array form. This means an attacker can deserialize malicious objects that should have been blocked by the security policy, potentially leading to arbitrary code execution during JSON/XML deserialization.

• CVE-2026-54591HIGH 8.1

  AsyncSSH, a Python library that enables secure shell connections, contains a flaw that allows a malicious SSH server to write files anywhere on a user's computer. When an AsyncSSH client connects to a compromised or hostile server, that server can trick the client into saving files outside the intended download directory by embedding path-traversal sequences (like '../') in filenames. This happens because the library doesn't properly validate where files are being written. The vulnerability affects all versions before 2.23.1 and requires user interaction—the user must initiate an SCP (secure copy) operation against an untrusted server.

• CVE-2026-54652HIGH 8.1

  Frigate version 0.17.1 contains a privilege escalation vulnerability affecting its log retrieval API. Any user with viewer-level access—the lowest privilege tier—can download application and web server logs that inadvertently contain sensitive credentials. These logs include auto-generated admin passwords and camera authentication details that may have been captured in web request parameters. This flaw enables a low-privilege user to escalate their access to administrator level by obtaining stored credentials from logs.

• CVE-2026-54771HIGH 8.1

  Langroid, a framework for building AI-powered chat applications, contains a vulnerability that allows attackers to bypass security controls and directly invoke tools even when those tools are explicitly disabled. An attacker with user-level access to a Langroid chat interface can send specially crafted JSON payloads to execute restricted functionality, leading to unauthorized actions. This affects versions prior to 0.65.3, and the issue has been resolved in that release.

• CVE-2026-54814HIGH 8.1

  StylemixThemes Motors contains a file inclusion vulnerability that allows an attacker to execute arbitrary PHP code on the affected server. The vulnerability stems from improper handling of filenames in PHP include/require statements, which can be exploited to load and execute malicious PHP files. An attacker does not need to authenticate or interact with a user to trigger this vulnerability, though the attack requires somewhat specific conditions to succeed. This is a serious flaw that can lead to full server compromise.

• CVE-2026-54842HIGH 8.1

  A missing authorization vulnerability in Royal Plugins' Royal MCP allows users with low-level login credentials to perform sensitive actions they shouldn't be allowed to perform. An attacker with a valid standard user account can exploit incorrectly configured access controls to gain unauthorized capabilities. The vulnerability affects Royal MCP versions up to and including 1.4.25. While authentication is required to exploit this issue, the low barrier to entry—needing only basic user credentials—makes this a meaningful risk for organizations using this plugin.

• CVE-2026-55119HIGH 8.1

  A vulnerability in Ubiquiti's UniFi Talk Application allows an attacker with basic network access and low-level user privileges to gain higher privileges within the application. An attacker would need to be authenticated to the network and application first, but once in, they can escalate their access level without additional user interaction. This type of privilege escalation is particularly concerning in enterprise environments where applications manage critical communications infrastructure.

• CVE-2026-55200HIGH 8.1

  libssh2 versions up to 1.11.1 contain a flaw that allows attackers to send malformed SSH packets over the network. The vulnerability fails to properly validate the size field in incoming packets, allowing an attacker to write data beyond the allocated memory buffer. This memory corruption can be leveraged to execute arbitrary code on systems running vulnerable versions. No authentication is required; an attacker anywhere on the network can trigger this vulnerability.

• CVE-2026-55388HIGH 8.1

  Piscina is a Node.js library that manages worker threads—a way to run code in parallel without blocking the main application. A vulnerability in how piscina handles configuration allows an attacker to inject malicious code by polluting JavaScript's Object.prototype, a shared object that all objects inherit from. When an attacker taints Object.prototype.filename before piscina tries to load a worker script, piscina will unknowingly use the attacker's malicious file path instead of the intended one, executing arbitrary code with the privileges of the worker thread. The vulnerability affects versions before 6.0.0-rc.2, 5.2.0, and 4.9.3.

• CVE-2026-55744HIGH 8.1

  Cotonti 1.0.0 contains a cross-site request forgery (CSRF) vulnerability in its Personal File Storage module that allows an attacker to trick authenticated users into uploading files to their account without their knowledge. The vulnerability exists because the file upload function fails to validate an anti-CSRF token that should prevent such unauthorized requests. An attacker can craft a malicious webpage that, when visited by a logged-in Cotonti user, silently triggers file uploads to that user's storage.

• CVE-2026-55762HIGH 8.1

  Rocket.Chat versions before 8.5.1 (and earlier maintenance branches) contain a critical authorization flaw in their cloud registration API endpoint. While the endpoint requires a user to be logged in, it fails to verify that the user has permission to perform workspace-level administrative actions. Any regular user can exploit this by sending a single API call to permanently disconnect their organization's Rocket.Chat instance from the cloud, destroying cloud credentials, revoking licenses, and disabling push notifications for all employees. Recovery requires manual intervention and re-registration with Rocket.Chat Cloud.

• CVE-2026-56020HIGH 8.1

  Webmin, a widely-deployed web-based system administration tool, contains a critical authentication bypass flaw in its HTTP server component. An attacker can forge an HTTP header to impersonate any user who has configured SSL client certificate authentication, gaining unauthorized access without valid credentials. The vulnerability affects Webmin versions prior to 2.641 and requires network access but does not require user interaction to exploit.

• CVE-2026-56076HIGH 8.1

  PraisonAI versions before 1.5.128 contain a vulnerability that allows attackers on the internet to remotely trigger agent execution and steal sensitive data without needing valid credentials. The flaw exists in the AGUI endpoint, which is misconfigured to accept requests from any origin and lacks proper authentication controls. An attacker can craft a request from a malicious website that, when visited by a user with network access to the vulnerable application, executes arbitrary agents and exfiltrates sensitive responses—including tool execution results and environment variables—directly to the attacker.

• CVE-2026-56123HIGH 8.1

  socat, a utility for establishing two-way network data streams, has a critical flaw in how it parses SOCKS5 proxy responses. When connecting through a malicious SOCKS5 proxy, an attacker can craft a specially formed response that causes socat to write arbitrary data into its own memory beyond the intended boundaries. This memory corruption could allow the attacker to crash the application or potentially execute code on the affected system. The vulnerability affects socat versions 1.8.0.0 through 1.8.1.1.

• CVE-2026-56243HIGH 8.1

  Capgo versions before 12.128.2 have a security control bypass flaw where the PostgREST/RLS plane accepts plaintext API keys even when the enforce_hashed_api_keys security setting is turned on. An authenticated attacker can send plaintext API keys directly to bypass the organization-level hashed-key enforcement and gain access to protected resources that should require stronger authentication.

• CVE-2026-56246HIGH 8.1

  Capgo versions before 12.128.2 have a critical flaw in how they handle API key permissions. When an administrator creates a restricted API key intended to work with only one organization, the system incorrectly allows that key to perform destructive actions on other organizations the administrator manages. This happens because the permission system checks the administrator's user privileges first, then fails to properly enforce the API key's scope restrictions. An attacker with legitimate API key access could delete organizations or remove members from organizations they shouldn't be able to touch.

• CVE-2026-56258HIGH 8.1

  Crawl4AI, a web scraping and automation framework, contains a flaw that allows attackers to write files anywhere on a system without authentication. The vulnerability exists in features that take screenshots and generate PDFs. Attackers can exploit this by manipulating file path parameters and using symlinks (shortcuts to files and directories) to bypass safety checks, potentially leading to system compromise if the application runs with elevated privileges.

• CVE-2026-56264HIGH 8.1

  Crawl4AI, a web scraping and automation tool, contains a critical flaw in versions before 0.8.7 that allows attackers to execute arbitrary JavaScript code on the server. The vulnerability exists in the /execute_js endpoint of the Docker API server, which was designed to run JavaScript but does so without proper security controls. Because the underlying browser runs with disabled security protections (--disable-web-security), an attacker can not only run malicious scripts but also abuse the server to attack internal systems on the network that it can reach—a technique known as server-side request forgery (SSRF).

• CVE-2026-56286HIGH 8.1

  Capgo, a mobile app framework, has a critical flaw in its account deletion feature that bypasses password verification. An attacker can delete any user's account without needing their password—by hijacking a session, exploiting a CSRF weakness, or manipulating request parameters. This leaves users vulnerable to account destruction and permanent data loss. The vulnerability affects Capgo versions prior to 12.128.2.

• CVE-2026-56345HIGH 8.1

  AVideo versions through 29.0 contain a critical flaw in the Meet plugin that allows attackers to impersonate any user, including administrators, without knowing their passwords. The vulnerability stems from the uploadRecordedVideo.json.php endpoint accepting user IDs directly from filenames without validation. An attacker who obtains the Meet shared secret—obtainable through separate path-traversal or timing attacks—can upload a specially crafted video file with a filename like '1-anything.mp4' to hijack any account. This is a complete authentication bypass that grants full system access.

• CVE-2026-56379HIGH 8.1

  ImageMagick, a widely-used open-source image processing library, contains a command injection vulnerability in how it handles SVG (Scalable Vector Graphics) files. An attacker can create a malicious SVG file that injects Magick Vector Graphics (MVG) commands, which are then executed during image rendering. This could allow an attacker to run arbitrary code on systems processing untrusted SVG files. The vulnerability affects ImageMagick versions before 7.1.2-15 (version 7.x branch) and before 6.9.13-40 (version 6.x branch).

• CVE-2026-56784HIGH 8.1

  OpenRemote versions before 1.25.0 allow authenticated users to delete alarm records belonging to other customers (tenants) by directly referencing those alarms' database IDs. An attacker with basic alarm-management permissions can guess or enumerate sequential alarm IDs and permanently erase alerts from separate organizations without authorization. The vulnerability stems from missing tenant-level access controls in the bulk deletion endpoint.

• CVE-2026-56876HIGH 8.1

  extract-zip, a popular Node.js package for unzipping files, fails to validate symlink targets during extraction. An attacker can craft a malicious zip archive containing symlinks with path traversal sequences (e.g., '../../../../etc/passwd') that point outside the intended extraction directory. When the library extracts these symlinks without validation, it creates links to arbitrary filesystem locations. Depending on the application's usage pattern and file permissions, this could enable an attacker to read sensitive files or overwrite critical data.

• CVE-2026-57751HIGH 8.1

  A CSRF vulnerability in Heateor Social Login version 1.1.39 and earlier allows attackers to trick authenticated users into performing unwanted actions without their knowledge. Unlike many CSRF flaws, this one does not require the attacker to be authenticated first—any malicious actor can craft a request that, when clicked or visited by a legitimate user, will execute actions on their behalf. The vulnerability is considered high-risk because it can lead to account compromise or unauthorized changes.

• CVE-2026-57950HIGH 8.1

  CVE-2026-57950 is a privilege-escalation vulnerability in ruoyi-vue-pro, an open-source ERP framework. The flaw allows users with shipment-level permissions (specifically those with 'erp:sale-out' access) to bypass intended access controls and perform unauthorized operations on sale orders—including viewing, creating, modifying, and deleting sensitive financial records. The vulnerability stems from a misconfiguration in the ErpSaleOrderController where the code checks for the wrong permission namespace, creating a gap between intended and actual authorization boundaries. This is particularly risky because it affects financially sensitive transactions and can be exploited without any user interaction or network complexity; a single authenticated request is sufficient.

• CVE-2026-5821HIGH 8.1

  The Image Optimizer plugin for WordPress contains a critical flaw that lets authenticated authors delete any file on the web server, not just their backup images. The vulnerability stems from the plugin storing file paths in a database field that users can edit directly, then using those paths without any safety checks when deleting attachments. An author-level user could inject a path to any system file they want removed, and the plugin will dutifully delete it when they remove an image. This affects all versions up to and including 1.7.4.

• CVE-2026-58282HIGH 8.1

  Microsoft Edge (Chromium-based) contains an access control flaw that allows attackers to spoof content over the network. An attacker does not need credentials or special privilege to exploit this vulnerability, though successful attacks require specific technical conditions to be met. The primary risk is that an attacker could impersonate trusted websites or services, potentially deceiving users into trusting malicious content.

• CVE-2026-58283HIGH 8.1

  Microsoft Edge (Chromium-based) contains a type confusion vulnerability that allows attackers to spoof content or resources over a network without requiring user interaction. The flaw arises from the browser accessing resources using incompatible data types, creating a logic gap that an unauthenticated attacker can exploit remotely. While the attack complexity is moderately high, successful exploitation could result in integrity compromise and limited information disclosure.

• CVE-2026-58286HIGH 8.1

  Microsoft Edge (Chromium-based) contains an access control flaw that allows attackers to spoof content or identity over the network without requiring user interaction or special privileges. The attacker must overcome some technical barriers to exploit it, but once successful, the impact on system integrity is significant. This is a HIGH severity issue affecting confidentiality, integrity, and availability across network boundaries.

• CVE-2026-58293HIGH 8.1

  Microsoft Edge (Chromium-based) contains a vulnerability that allows an attacker to manipulate how files are named or accessed on your system, potentially leading to code execution without requiring the user to do anything special. The vulnerability is network-reachable and carries high severity due to its potential for complete system compromise.

• CVE-2026-58370HIGH 8.1

  Woodpecker, a CI/CD orchestration platform, has a critical authentication bypass affecting its GitLab integration. When a developer opens a merge request from a forked repository, Woodpecker checks whether the pipeline requires approval by comparing the commit author name against a whitelist. On GitLab, this author name comes directly from the git commit metadata supplied in the webhook—metadata that any user can forge. An attacker can craft a commit with an author name matching an approved user, tricking Woodpecker into skipping the mandatory approval gate. The pipeline then executes the attacker's code on Woodpecker agents with access to CI secrets. This affects only GitLab; other supported forges (GitHub, Gitea, Forgejo, Bitbucket) use cryptographically signed identity information and are not vulnerable.

• CVE-2026-58372HIGH 8.1

  SeaweedFS, a distributed file system and cloud storage solution, contains a critical authorization bypass flaw in its S3-compatible gateway. An authenticated user with write access to one S3 bucket can exploit a path traversal weakness to delete files in completely different buckets belonging to other tenants. The vulnerability stems from the system validating bucket authorization at the URL level but failing to inspect the contents of deletion request bodies, where attackers can inject directory-traversal sequences (../) to escape the intended bucket boundary and reach objects they should not be able to access.

• CVE-2026-58377HIGH 8.1

  JeecgBoot, a Java-based rapid application development framework, contains a critical flaw in how it controls access to API credential management. Low-privilege authenticated users can view, create, modify, and delete all OpenAPI credentials (API keys and secrets) without proper permission checks. This is particularly dangerous because the system returns secret keys in plaintext, allowing attackers to steal credentials and impersonate legitimate API clients.

• CVE-2026-59712HIGH 8.1

  Leantime, a project management platform, has a vulnerability in its JSON-RPC API that allows any authenticated user to view sensitive information about other users—including password hashes, two-factor authentication secrets, and active session tokens. An attacker with valid login credentials can request user data for any account ID and systematically extract credentials across the entire user base, enabling offline password attacks, 2FA circumvention, and session takeover without needing administrator access.

• CVE-2026-59713HIGH 8.1

  Leantime's OIDC (OpenID Connect) single sign-on implementation contains a critical flaw in how it validates login callbacks. When users are redirected back from an identity provider after authentication, Leantime fails to properly verify that the response came from a legitimate request. An attacker can trick a victim into clicking a malicious link that logs them into Leantime using the attacker's own credentials, effectively hijacking the victim's session. This is a session fixation attack that requires user interaction but poses a serious risk to multi-user Leantime deployments, especially those relying on OIDC for centralized authentication.

• CVE-2026-59795HIGH 8.1

  JetBrains TeamCity versions prior to 2026.1.2 contain a stored cross-site scripting (XSS) vulnerability in the unauthenticated agent registration process. An attacker without authentication can inject malicious scripts during agent registration, which are then stored and executed in the browsers of users who interact with the affected TeamCity instance. This allows attackers to steal session tokens, perform actions on behalf of administrators, or redirect users to malicious sites.

• CVE-2026-59796HIGH 8.1

  JetBrains TeamCity versions before 2026.1.2 contain a flaw that allows authenticated users to modify CI/CD pipelines without proper authorization checks. An attacker with valid login credentials can bypass permission controls to alter pipeline configurations, potentially injecting malicious build steps, exfiltrating secrets, or compromising downstream deployments. This is a privilege escalation vulnerability affecting the integrity of your build and release process.

• CVE-2026-6075HIGH 8.1

  The Media Library Assistant plugin for WordPress contains a Cross-Site Request Forgery (CSRF) vulnerability affecting versions up to 3.35. The flaw allows attackers to craft malicious web pages or emails that, when viewed by an administrator, can trigger unauthorized bulk operations on plugin settings and attachment metadata without the admin's knowledge or consent. An attacker doesn't need valid WordPress credentials to exploit this—only the ability to trick an admin into visiting a compromised site.

• CVE-2026-6250HIGH 8.1

  A flaw in the ONVIF service of TP-Link Tapo C110 v2 cameras allows an authenticated attacker to trigger a factory reset by exploiting how the device processes user input. When the device interprets attacker-supplied data as a format string—a type of code injection—an attacker can manipulate memory to redirect the camera to reset itself, erasing all saved configurations, login credentials, and effectively disabling the device until reconfiguration.

• CVE-2026-6455HIGH 8.1

  WP Contact Form 7 DB Handler, a popular WordPress plugin for managing contact form submissions, contains a critical vulnerability that allows attackers to delete arbitrary files from affected WordPress servers. An attacker can trick an administrator into visiting a malicious webpage, which silently executes commands to remove critical system files—potentially crippling the entire website. The vulnerability chains together three separate weaknesses: a bypassable security check, improper handling of user input in database queries, and unsafe processing of untrusted data. Versions 3.0 and earlier are affected.

• CVE-2026-7311HIGH 8.1

  The TinyPNG compression plugin for WordPress contains a dangerous flaw that lets authenticated users with author-level permissions delete any file on the web server. An attacker exploits this by manipulating image metadata during the attachment deletion process, allowing them to target critical files like wp-config.php. Deleting the right file can give the attacker full control of the WordPress installation. The vulnerability affects all versions up to and including 3.6.13.

• CVE-2026-7368HIGH 8.1

  The Yarbo cloud platform lacks proper authorization controls, allowing any authenticated user—whether using shared credentials or legitimate accounts—to monitor all robots globally and send commands to any robot if they know its serial number. The serial number is routinely exposed in telemetry data, making it discoverable. This means a single compromised credential, even after hard-coded credentials are removed, grants an attacker complete control over an entire fleet of robots without per-device or per-user access restrictions.

• CVE-2026-7383HIGH 8.1

  OpenSSL contains a flaw in how it calculates memory buffer sizes when converting text strings to Unicode formats. When processing very large inputs—on the order of half a gigabyte or more—the size calculation wraps around to an incorrect (much smaller) value due to signed integer overflow. This causes OpenSSL to allocate a tiny buffer but then write far more data into it, corrupting the heap. An attacker who can supply specially crafted input directly to the affected functions could crash the application or potentially execute arbitrary code, though real-world exploitation requires unusual circumstances.

• CVE-2026-7656HIGH 8.1

  A logic error in the Zephyr IPv6 Neighbor Discovery code allows attackers to bypass critical validation checks by exploiting an incorrect boolean operator precedence. When legitimate Neighbor Discovery messages arrive with ICMPv6 code 0 (the normal value), the buggy condition evaluates false, causing the packet to skip all security checks entirely. An adjacent on-link attacker can send forged Router Advertisements to hijack routing, reconfigure DNS, and adjust network timers, or craft fake Neighbor messages to poison the neighbor cache and intercept traffic. Because the Hop Limit 255 verification is bypassed, a sophisticated remote attacker may also exploit this from off-link in certain network topologies. The flaw is purely a validation logic error—not a memory bug—since the underlying packet-parsing functions themselves are bounds-safe.

• CVE-2026-8095HIGH 8.1

  A WordPress plugin called Frontend File Manager contains a security flaw that allows logged-in users with basic subscriber accounts to delete critical files from a website's server. The vulnerability exists because the plugin fails to properly validate file paths when processing deletion requests, enabling attackers to specify any file on the system for removal—including the configuration file that stores database credentials. This could allow a low-privileged attacker to completely disable or take over a website. The flaw affects all versions up to 23.6.

• CVE-2026-8147HIGH 8.1

  MLflow, a popular machine learning tracking and versioning platform, has a critical authorization bypass vulnerability in its trace API endpoints. When authentication is enabled, authenticated users can access, modify, or delete machine learning experiment traces they shouldn't be able to see—essentially breaking the permission boundaries that organizations set up to isolate experiments. An attacker with any valid login can read sensitive training data, destroy audit records, or tamper with experiment metadata across the entire platform. The flaw stems from missing authorization checks in the trace API handler, allowing requests to skip permission validation entirely.

• CVE-2026-8286HIGH 8.1

  A vulnerability in curl allows a new network transfer to incorrectly reuse an existing active connection when upgrading to TLS encryption via STARTTLS, even when the TLS settings don't match between the two transfers. This connection reuse could allow an attacker to intercept or manipulate traffic that should be encrypted with different security parameters, compromising the confidentiality and integrity of data in transit.

• CVE-2026-8442HIGH 8.1

  The WP Review Slider Pro WordPress plugin contains a file deletion vulnerability that allows attackers with subscriber-level account access to delete arbitrary files from the web server. The flaw stems from incomplete authorization checks and insufficient validation of file paths in the plugin's AJAX handlers. An attacker could exploit this to remove critical site files, potentially creating conditions for further compromise such as remote code execution. The vulnerability affects all versions up to and including 12.6.8.

• CVE-2026-8796HIGH 8.1

  Sereal::Decoder, a Perl serialization library, contains a heap memory safety flaw that allows attackers to read sensitive data from memory. By crafting malicious serialized input, an attacker can trick the decoder into reading beyond the intended buffer boundaries, potentially exposing heap memory contents. The vulnerability requires user interaction (such as processing an attacker-supplied file or message) but does not require special privileges and can be triggered remotely.

• CVE-2026-9072HIGH 8.1

  IBM WebSphere Application Server and WebSphere Application Server Liberty face a critical remote code execution vulnerability when Intelligent Management is enabled with the WebServer Plug-in component. An attacker who can intercept or impersonate backend server communications can send malicious responses that execute arbitrary code on the affected application server or cause it to crash. The attack requires network access but no user interaction or authentication, making it a serious threat in production environments.

• CVE-2026-9095HIGH 8.1

  Casdoor, an open-source identity and access management platform, contains a critical flaw in how it processes SAML single sign-on requests. When a user logs in via SAML, the application accepts and authenticates the SAML assertion without checking whether it has already been used. An attacker who intercepts or obtains a valid SAML assertion—even days or weeks after it was originally issued—can replay it to gain a logged-in session as that user. This bypasses password and multi-factor authentication requirements entirely. The risk is particularly severe if an attacker targets administrator accounts, as they could gain full control of the Casdoor instance.

• CVE-2026-9213HIGH 8.1

  NETGEAR gaming routers are vulnerable to remote code execution if an attacker can intercept and modify network traffic between your router and the internet. The vulnerability affects several popular NETGEAR gaming models. While the attack requires the ability to position oneself in the network path (man-in-the-middle capability), successful exploitation grants the attacker complete control over the router, potentially compromising your entire home or office network.