By severity

Medium-severity vulnerabilities

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

4010 published vulnerabilities · page 8 of 41

• CVE-2026-50519MEDIUM 6.5

  GitHub Copilot Chat is vulnerable to unauthorized information disclosure due to insecure default initialization settings. An attacker can exploit this over the network to retrieve sensitive information, though they require user interaction to succeed. The vulnerability affects default configurations and does not allow modification or deletion of data.

• CVE-2026-50630MEDIUM 6.5

  A flaw in Apache CXF's OAuth2 authentication mechanism allows attackers to inject line breaks into HTTP response headers. By crafting a malicious realm parameter, an attacker can trick the server into sending arbitrary headers or fragmenting the response entirely, potentially leading to session hijacking, cache poisoning, or other header-based attacks. The vulnerability affects versions before 4.2.2 and 4.1.7.

• CVE-2026-50634MEDIUM 6.5

  Apache CXF's JwsJsonContainerRequestFilter contains a flaw that allows attackers to bypass signature verification on request metadata. An attacker can send a request with tampered HTTP headers or Content-Type information that the application believes came from a verified signature, but actually did not. This mismatch between what the filter validates and what downstream application code processes can lead to security assumptions being violated, potentially enabling unauthorized data access or modification.

• CVE-2026-50639MEDIUM 6.5

  A Perl metrics reporting library fails to validate input containing special characters, allowing attackers to inject fake metrics into monitoring systems. The vulnerability affects Metrics::Any::Adapter::SignalFx versions before 0.04, which extends statsd protocol support but does not properly filter newlines and control characters in metric labels. An attacker could craft requests with embedded metric definitions to pollute monitoring data, potentially masking real system issues or triggering false alerts.

• CVE-2026-5074MEDIUM 6.5

  ARMember Premium, a WordPress plugin, contains a SQL Injection vulnerability in its AJAX handler that processes user-supplied sort parameters without proper validation. An authenticated user with basic Subscriber privileges or higher can craft malicious input to extract sensitive data from the site's database. The vulnerability only impacts sites that have explicitly enabled the optional 'User Private Content' addon, which is disabled by default, limiting the blast radius.

• CVE-2026-50811MEDIUM 6.5

  FreeType, a widely-used font rendering library, contains an out-of-bounds read flaw in its variable font handling code. When processing specially crafted TrueType variable fonts, the library reads memory beyond allocated boundaries, potentially exposing sensitive data or causing the application to crash. This affects FreeType version 2.14.3 and earlier versions prior to a specific commit that patches the issue.

• CVE-2026-50892MEDIUM 6.5

  Nginx Proxy Manager version 2.14.0 contains a flaw in how it controls access to its Let's Encrypt certificate download feature. An authenticated user can bypass the intended access controls and retrieve TLS private keys that should remain confidential. This requires the attacker to already have valid credentials to the system, but once authenticated, they can extract sensitive cryptographic material via a specially crafted web request.

• CVE-2026-5135MEDIUM 6.5

  Foreman, a popular open-source infrastructure management platform used by Red Hat Satellite and Red Hat Enterprise Linux environments, contains an access control flaw that allows authenticated users to abuse their host-editing privileges in an unintended way. Specifically, a user with permission to edit one host's configuration can manipulate lookup value overrides—settings that customize how hosts behave—and reassign them to different hosts that the user shouldn't have authority to modify. This is done by exploiting how the system processes nested host attributes when validating which configuration should apply to which system. The result is potential unauthorized changes to host configurations across organizational and location boundaries, meaning a user could alter critical settings on systems outside their intended scope of access.

• CVE-2026-5142MEDIUM 6.5

  Foreman, an infrastructure lifecycle management tool, contains an authorization flaw that allows authenticated users to access SSH private keys belonging to other organizations. Users with 'view_keypairs' permission can bypass organization-level scoping by directly requesting specific key pair IDs, exposing sensitive credentials across tenant boundaries. This is a multi-tenant isolation failure that affects deployments where multiple organizations share a Foreman instance.

• CVE-2026-5149MEDIUM 6.5

  A flaw in the RTMKit WordPress plugin allows users with basic contributor-level access or higher to view form submissions that belong to other users. An attacker can exploit this by guessing or iterating submission IDs in an AJAX request that doesn't properly check permissions. This affects all versions up to 2.0.7 and exposes sensitive form data without additional barriers once an attacker gains even modest WordPress account privileges.

• CVE-2026-51598MEDIUM 6.5

  A vulnerability in MERCURY MIPC252W IP cameras allows an attacker on the local network to crash the device by sending a specially crafted video streaming request. The camera's RTSP (Real-Time Streaming Protocol) service fails to properly validate the request format, and a malformed URL in the request line triggers a denial of service condition. No authentication is required—an attacker simply needs network access to the camera. This is a stability issue rather than a data breach or unauthorized access threat, but it can disrupt surveillance operations.

• CVE-2026-51946MEDIUM 6.5

  GoAdmin, a web administration framework, contains a SQL injection flaw in its table information endpoints. Attackers can manipulate the __sort_type URL parameter to inject SQL commands, potentially reading sensitive database records or modifying data. The vulnerability affects all versions up to and including v1.2.26 and requires no authentication to exploit.

• CVE-2026-52188MEDIUM 6.5

  A buffer overflow flaw in UTT nv518G devices running firmware version nv518GV3v3.2.7-210919-161313 can be triggered remotely to crash the device or disrupt its operations. An attacker on the local network can send specially crafted network packets to the gohead HTTP service component, causing the device to run out of memory or fail, resulting in a denial-of-service condition. This is a memory safety issue where input validation is insufficient to prevent writing beyond allocated buffers.

• CVE-2026-52673MEDIUM 6.5

  Cboard versions 0.4.2 and earlier contain a SQL injection flaw in the getDimensionsValues component that allows authenticated attackers to extract sensitive data from the application database. The vulnerability requires login credentials but does not require user interaction once authenticated, making it a practical concern for organizations deploying Cboard with untrusted or compromised user accounts.

• CVE-2026-52718MEDIUM 6.5

  GStreamer, a widely-used multimedia framework, contains a bug in its AV1 video codec parser that causes it to misinterpret data and crash when processing certain malformed video files. An attacker could distribute a specially crafted AV1 video file that, when opened by a user, immediately crashes the application handling it. While this doesn't allow data theft or system compromise, it enables denial of service—disrupting legitimate media playback and potentially affecting workflows that depend on reliable video processing.

• CVE-2026-52866MEDIUM 6.5

  A nearby attacker can flood a device's Bluetooth Low Energy (BLE) connection slot, locking out legitimate users and applications from connecting. The attacker doesn't need credentials or user interaction—they simply need to be within wireless range. This is a denial-of-service attack that affects availability rather than confidentiality or integrity.

• CVE-2026-53466MEDIUM 6.5

  ImageMagick, a widely-used image editing and manipulation library, contains an integer overflow flaw in its XCF (GIMP native format) decoder. When processing a specially crafted XCF file, the overflow can trigger an out-of-bounds memory read, potentially crashing the application. The vulnerability affects ImageMagick versions before 6.9.13-51 and 7.1.2-26. While the immediate risk is denial of service through application crashes, the out-of-bounds read could theoretically expose sensitive information from adjacent memory regions.

• CVE-2026-53489MEDIUM 6.5

  A vulnerability in containerd, the widely-used container runtime powering Kubernetes and Docker environments, allows an authenticated user to read arbitrary files on the host system. The flaw exists in how the CRI plugin (Container Runtime Interface) handles container logs when restoring from checkpoint images—it fails to validate symlink paths, enabling an attacker to redirect log requests to sensitive files elsewhere on the system. An attacker with local access or pod-level permissions could exploit this via kubectl logs commands to extract confidential data. The issue affects containerd versions prior to 2.3.2, 2.2.5, and 2.1.9.

• CVE-2026-53520MEDIUM 6.5

  Nezha Monitoring, a self-hosted server and website monitoring tool, contains a privilege escalation flaw affecting versions 2.0.14 through 2.0.x. Any authenticated user can claim ownership of the monitoring dashboard's host configuration through a NAT-based mechanism, effectively taking control of all dashboard routing. This allows an internal attacker with valid credentials to hijack the monitoring infrastructure's core routing and potentially lock out legitimate administrators. The vulnerability has been resolved in version 2.1.0.

• CVE-2026-53522MEDIUM 6.5

  Nezha Monitoring, a self-hosted server monitoring tool, contains a resource exhaustion vulnerability in versions 1.0.0 through 2.1.x. Two API endpoints that establish persistent WebSocket connections to monitored servers fail to implement any limits on how many concurrent connections a single authenticated user can create. An attacker with valid credentials can repeatedly call these endpoints to exhaust server memory and crash the monitoring dashboard, disrupting visibility into your infrastructure.

• CVE-2026-53577MEDIUM 6.5

  Kestra, an open-source workflow orchestration platform, has an access control flaw in its file preview feature that allows any logged-in user to read output files from executions they shouldn't have access to. The vulnerability bypasses protections designed to isolate work between different execution runs and namespaces within the same tenant. An attacker with valid credentials—even a low-privileged user—can request files from any other execution and retrieve sensitive data. Versions 1.0.45 and 1.3.21 patch this issue.

• CVE-2026-53698MEDIUM 6.5

  Silverpeas, a collaborative platform software, has a flaw in how it handles the 'Personal space' feature when no component ID is provided. An authenticated attacker can exploit this misconfiguration to view sensitive files or data in personal workspace areas that should not be accessible to them. The vulnerability requires valid user credentials to exploit but does not require user interaction, making it a practical concern in multi-tenant or multi-user Silverpeas deployments.

• CVE-2026-53701MEDIUM 6.5

  A flaw exists in GStreamer's H.266/VVC video codec parser that allows an attacker to write data beyond the boundaries of memory buffers. When processing specially crafted video files, the parser fails to properly validate loop boundaries while organizing video slices, potentially writing to unintended memory locations. This can cause the application to crash or behave unexpectedly. The vulnerability requires user interaction (opening or playing a malicious video file) but can be triggered remotely by distributing the crafted file.

• CVE-2026-53702MEDIUM 6.5

  A flaw in GStreamer's H.265 video codec parser allows a specially crafted video file to trigger a crash or memory corruption on systems that process it. The bug occurs when the parser mishandles internal buffer size calculations during video stream analysis, writing data beyond allocated memory on the stack. An attacker could distribute a malicious video file that crashes applications or services using this library, but direct code execution is not indicated by the vulnerability mechanics.

• CVE-2026-53808MEDIUM 6.5

  OpenClaw versions prior to 2026.5.6 contain a bypass in their Skill Workshop approval controls. An attacker can force workshop changes to be applied immediately—bypassing the system's built-in approval requirement—by directly calling an agent tool with apply: true. This means configuration modifications intended to require manager review can slip through without authorization. The vulnerability requires user interaction (clicking a link or opening a message) but no special privileges to trigger.

• CVE-2026-53815MEDIUM 6.5

  OpenClaw versions before 2026.5.19 allow authenticated users with limited privileges to read messages from channels they should not have access to. The vulnerability stems from missing validation checks that would normally restrict users to an allowlist of permitted channels. An attacker with any login credentials could exploit this to view sensitive communications intended only for specific teams or roles.

• CVE-2026-53824MEDIUM 6.5

  OpenClaw versions before 2026.4.24 have a token revocation flaw that briefly keeps certain user accounts active even after their access should have been cut off. When administrators revoke user tokens—the credentials that grant permission to run automated commands—a window of time can exist where those revoked users can still execute commands while the system refreshes its security checks. An attacker with a revoked token could exploit this gap to run unauthorized actions, with the actual impact depending on what command permissions the operator has configured for that token.

• CVE-2026-53825MEDIUM 6.5

  OpenClaw versions prior to 2026.4.7 contain a file-reading vulnerability in the memory-wiki ingest feature. Users with operator.write permissions can exploit this to read files from the local system that fall outside the intended ingest sources, potentially accessing sensitive configuration files, credentials, or other protected data. The vulnerability requires authenticated access and operator-level permissions, limiting its reach but increasing risk for multi-tenant or federated deployments where gateway operators may be less trusted.

• CVE-2026-53827MEDIUM 6.5

  OpenClaw versions prior to 2026.5.2 contain a credential exposure flaw in how the message.action forwarding mechanism handles metadata. An authenticated attacker can manipulate model-controlled metadata to trick the system into forwarding sensitive action payloads—including Gateway credentials—to attacker-controlled loopback URLs. By doing so, the attacker intercepts authentication tokens and action data without modifying system behavior or causing outages.

• CVE-2026-53830MEDIUM 6.5

  OpenClaw versions before 2026.4.22 contain a vulnerability in how webhook secrets are revoked. When an administrator reloads secrets to revoke old Slack and Zalo webhook credentials, those old secrets may remain functional for a window of time. An attacker who obtained a revoked webhook secret could continue to use it to deliver webhook events even after the operator believes the secret has been invalidated. This creates a race condition where stale credentials continue to work when they shouldn't.

• CVE-2026-53839MEDIUM 6.5

  OpenClaw versions before 2026.5.7 contain a flaw in how they validate hostnames when processing retry requests. Instead of checking for an exact match to trusted hosts, the software accepts any hostname that starts with the trusted hostname prefix. An authenticated attacker can exploit this by creating a crafted hostname that begins with a legitimate server name—for example, turning 'trusted.com' into 'trusted.com.evil.net'—and trick the system into sending sensitive authentication credentials to their malicious endpoint.

• CVE-2026-53844MEDIUM 6.5

  OpenClaw versions before 2026.4.29 contain a flaw that allows authenticated users to bypass session visibility controls when searching shared memory. An attacker with valid credentials can craft searches to retrieve memory entries that should be restricted from their session, exposing data they should not have access to. This is a data exposure risk that requires valid authentication to exploit, but once inside the system, attackers can circumvent intended data compartmentalization.

• CVE-2026-53854MEDIUM 6.5

  OpenClaw versions before 2026.4.25 have a privilege escalation flaw in how they handle authentication for internal and webchat commands. An authenticated attacker can exploit wildcard permission rules that incorrectly carry over between separate communication channels, allowing them to run commands with owner-level privileges in areas where they shouldn't have that access. This breaks the intended isolation between different parts of the system.

• CVE-2026-53859MEDIUM 6.5

  OpenClaw versions before 2026.5.26 have a hostname validation flaw that allows authenticated attackers to bypass network access controls. The vulnerability exploits how the application validates hostnames in URLs derived from model or workspace settings. By appending a trailing dot to a hostname (e.g., "blocked-host.com." instead of "blocked-host.com"), attackers can circumvent blocklist policies that administrators set up to restrict which systems OpenClaw can reach. This is a logic error, not a cryptographic weakness, and requires existing user access to exploit.

• CVE-2026-53899MEDIUM 6.5

  Firefox for iOS versions prior to 152.0 contained a cookie-handling flaw that could allow attackers to steal session credentials. When a user visited a malicious website, that site could intercept cookies meant for a legitimate target domain if the malicious site's domain was a suffix match (for example, attacker.example.com could receive cookies for example.com). The vulnerability specifically affects PDF request handling, where the browser's cookie attachment logic used incomplete domain validation. An attacker would need to trick a user into visiting their malicious site while the user had an active session on the target domain.

• CVE-2026-53902MEDIUM 6.5

  MCO (MyComplianceOffice) contains a flaw that allows authenticated users to bypass authorization controls and add themselves to groups they shouldn't access. An attacker with valid login credentials could escalate their privileges by joining arbitrary groups, potentially gaining unauthorized access to sensitive functions or data restricted to those groups. The vulnerability requires authentication and was confirmed in version 25.3.3.1, though other versions may be affected.

• CVE-2026-53909MEDIUM 6.5

  MCO (MyComplianceOffice) contains a file upload vulnerability where the application fails to validate file types on the server side. An attacker with legitimate user credentials can bypass client-side restrictions and upload arbitrary file types to the system. This could allow malicious files such as executables, scripts, or other dangerous content to be stored on the server, potentially leading to further compromise depending on how uploaded files are processed or served.

• CVE-2026-53961MEDIUM 6.5

  Discourse instances running before specific patched versions are vulnerable to a webhook validation bypass affecting AWS Simple Email Service (SES) integration. An attacker with any AWS account can send forged bounce notifications to a target Discourse instance, causing legitimate user email addresses to be revoked from the platform. The flaw stems from incomplete validation of AWS SNS messages—while Discourse verified the cryptographic signature, it failed to confirm that bounce notifications came from the expected AWS topic, leaving the door open for spoofing.

• CVE-2026-53982MEDIUM 6.5

  Cap-go Console versions before 12.28.2 contain a flaw that allows attackers with valid account credentials to trigger a denial-of-service condition affecting device authentication. When an attacker initiates account deletion while a device identifier is active in a session, the platform incorrectly locks that device to a disabled-account page for approximately 30 days. This prevents legitimate users from logging in or creating new accounts on that device or browser, even if they own the account or are attempting to register fresh.

• CVE-2026-54007MEDIUM 6.5

  Open WebUI versions prior to 0.9.6 contain a cross-origin messaging vulnerability that allows an attacker-controlled website to forge chat requests on behalf of an authenticated user. When a victim visits a malicious site while logged into their Open WebUI instance, that site can silently inject prompts and trigger API calls to create new chats or request completions—all without the user's knowledge or consent. The attacker effectively hijacks the user's AI session to execute queries or actions under the victim's credentials.

• CVE-2026-54009MEDIUM 6.5

  Open WebUI, a self-hosted AI platform, contains an authorization flaw that allows authenticated users to read files belonging to other users. When uploading images to the chat API, the system fails to verify file ownership before processing. An attacker can reference another user's file ID, causing the server to retrieve that file, encode it, and pass it to the language model—effectively using the LLM as a file reader to exfiltrate private content. The issue was introduced before version 0.9.6 and has been patched in that release.

• CVE-2026-54019MEDIUM 6.5

  Open WebUI, a self-hosted AI platform designed for offline operation, contains a vulnerability in how it controls access to data collections when running in Milvus multitenancy mode. Even though version 0.9.6 added access controls at the collection level, an attacker with valid credentials can bypass these protections by exploiting how user-supplied collection names are processed. The vulnerability allows an authenticated user to access or manipulate collections they should not have permission to view, potentially exposing sensitive AI training data or model information. This is a partial fix for an earlier vulnerability (CVE-2026-44560) that was incompletely addressed.

• CVE-2026-54024MEDIUM 6.5

  LibreChat, a ChatGPT alternative supporting multiple AI providers, contains a file upload vulnerability in its conversation import endpoint. An authenticated user can upload arbitrarily large files to the server, consuming disk space and memory until the service becomes unavailable. This occurs because the import endpoint uses a separate file upload handler that wasn't updated when size limits were added elsewhere in the application, and the default configuration leaves the size check disabled. The vulnerability affects versions prior to 0.8.4-rc1.

• CVE-2026-54027MEDIUM 6.5

  LibreChat, a multi-provider AI chat application, contains an authorization bypass in its image upload endpoint. Authenticated users can upload files into any agent's resource storage without permission checks, bypassing existing controls on the standard file upload route. An attacker with basic login credentials can inject malicious files into other users' agents, potentially compromising their AI tool execution environments.

• CVE-2026-54037MEDIUM 6.5

  LibreChat, an open-source ChatGPT alternative that integrates multiple AI providers, contains a rate-limiting bypass vulnerability. A previous security fix (CVE-2025-7105) added protections to one conversation-duplication endpoint, but an identical endpoint was overlooked. Authenticated users can exploit the unprotected endpoint to trigger the same resource-exhaustive operations, circumventing the earlier mitigation and potentially causing denial of service.

• CVE-2026-54092MEDIUM 6.5

  File Browser versions prior to 2.63.6 contain a denial-of-service vulnerability in the login API that accepts arbitrarily large passwords without validation. An authenticated attacker can submit an extremely long password string, causing the application to consume excessive CPU and memory resources, leading to service degradation or complete failure. In containerized environments, this can crash the container, severely impact host system performance, and potentially trigger errors in the Docker daemon itself.

• CVE-2026-54164MEDIUM 6.5

  API Platform, a popular framework for building REST and GraphQL APIs, contains a type-confusion vulnerability in how it processes related resources. When an API accepts write requests (creating, updating, or patching data), an authenticated attacker can supply a link to a resource of the wrong type, and the system may silently accept it instead of rejecting the mismatch. This corrupts the internal integrity of the data model. The vulnerability affects versions before 4.1.30, 4.2.26, and 4.3.12, though the impact varies depending on whether the API uses older or newer PHP coding patterns.

• CVE-2026-54233MEDIUM 6.5

  vLLM, an open-source inference engine for large language models, contains a denial-of-service vulnerability in its audio transcription endpoint. The flaw allows an authenticated user to upload a heavily compressed audio file that expands dramatically upon decoding, consuming excessive server memory and crashing the service. A 25 MB compressed OPUS file can balloon to approximately 14.9 GB of uncompressed audio data, overwhelming the system. This affects vLLM versions prior to 0.23.1rc0.

• CVE-2026-54235MEDIUM 6.5

  vLLM, a popular engine for running and serving large language models, has a flaw in how it validates temperature parameters before sending them to GPU processing. When users supply invalid temperature values—specifically NaN (not-a-number) or positive infinity—the validation checks incorrectly pass them through because Python's comparison operators don't handle these special float values as expected. Those malformed values then reach the GPU sampling kernels, causing crashes or undefined behavior that disrupts inference workloads. The issue affects versions before 0.23.1rc0 and is resolved in that release.

• CVE-2026-54261MEDIUM 6.5

  Wagtail, a Django-based open-source content management system, contains a missing permission check on its image preview endpoint. An admin user with access to Wagtail's administrative interface can preview any image stored in the system without proper authorization controls. However, the vulnerability does not expose the underlying image metadata or object data itself, and it cannot be exploited by regular website visitors—only by someone who already has admin credentials. The issue affects versions before 7.0.8, 7.3.3, and 7.4.2 and has been patched in those releases.

• CVE-2026-54288MEDIUM 6.5

  Hono is a popular JavaScript framework used to build web applications across different runtime environments. Versions before 4.12.25 contain a flaw in how they enforce request body size limits on AWS Lambda. An attacker can bypass the body size check by declaring a small content length in the HTTP header while actually sending a much larger payload. This allows oversized requests to reach the application, potentially causing denial of service or resource exhaustion. The issue has been patched in version 4.12.25.

• CVE-2026-54324MEDIUM 6.5

  Daytona, an AI code execution platform, contained a cross-tenant authorization weakness in its notification system. Before version 0.185.0, any authenticated user could subscribe to real-time event channels belonging to other organizations and passively observe their notifications. An attacker with valid credentials could monitor sensitive operational events from competing or unrelated tenants without detection. The vulnerability does not allow data modification or service disruption, but it enables unauthorized information disclosure across organizational boundaries.

• CVE-2026-54448MEDIUM 6.5

  Trivy, a widely-used container and artifact security scanner, contains a denial-of-service vulnerability in its Helm chart processing logic. When Trivy scans a malicious Helm chart archive (.tgz file), it decompresses the contents without enforcing memory limits, allowing an attacker to craft a small file that expands to gigabytes in memory. This causes the Trivy process to consume excessive RAM and be killed by the operating system, disrupting security scanning operations. The vulnerability affects versions prior to 0.71.0 and requires the attacker to have the ability to place a malicious .tgz file where Trivy will scan it.

• CVE-2026-54468MEDIUM 6.5

  Dell Unisphere for PowerMax versions 10.3.0.5 and earlier contain a flaw that allows an authenticated attacker with basic network access to bypass file path restrictions and read files they should not be able to access. The vulnerability requires valid credentials but does not need user interaction to trigger, making it a concern for organizations with untrusted internal users or compromised service accounts.

• CVE-2026-54518MEDIUM 6.5

  Jackson-databind, a widely-used Java library for converting JSON to objects, has a flaw where constructor parameters can be populated with JSON data even when access controls (views) should prevent it. Specifically, when a constructor parameter is marked both @JsonUnwrapped and @JsonView(AdminView.class), the library bypasses the view restriction during JSON deserialization, allowing an attacker to inject data into admin-only fields. This affects versions 2.21.0 through 2.21.3 and 3.1.3, and is patched in 2.21.4 and 3.1.4.

• CVE-2026-54704MEDIUM 6.5

  OpenTelemetry Java Instrumentation versions before 2.28.0 fail to properly mask database passwords in SQL CONNECT statements when those passwords are enclosed in double quotes. This oversight causes plaintext credentials to leak into distributed trace spans that are then exported to observability platforms like Datadog, New Relic, or Jaeger. An authenticated attacker or insider with access to those observability backends could extract database credentials, potentially gaining unauthorized database access. The flaw affects the JDBC auto-instrumentation component, which automatically captures database connection metadata for monitoring.

• CVE-2026-54775MEDIUM 6.5

  CoreWCF, a .NET Core implementation of Windows Communication Foundation (WCF), contains a denial-of-service vulnerability in versions prior to 1.8.1 and 1.9.1. When a Kafka-based CoreWCF service receives a null-value tombstone record—a special Kafka message type used to signal deletion—the service stops processing all subsequent messages from that topic. An attacker with permission to produce messages to the Kafka topic can exploit this to disable the affected endpoint indefinitely, forcing service restarts or intervention to recover. This is a network-accessible vulnerability requiring only standard produce credentials.

• CVE-2026-54777MEDIUM 6.5

  CoreWCF, a .NET Core implementation of Windows Communication Foundation, contains a local privilege escalation and eavesdropping vulnerability in its NetNamedPipe transport layer. Attackers with local access can intercept Named Pipe traffic by attaching to a pre-existing pipe instance during a narrow window when the service is starting up. This race condition occurs between the moment the service publishes its shared memory GUID and when it actually creates the named pipe, allowing a local attacker to position themselves in the middle of communications and read or modify traffic.

• CVE-2026-54798MEDIUM 6.5

  CPCI85 and SICORE Base system versions before 26.20 contain an exposed debugging interface accessible over the network that can be reached by authenticated users. An attacker with valid credentials could exploit this interface to crash the web process, causing the system to become unavailable. This is a denial-of-service vulnerability affecting industrial control and communication systems.

• CVE-2026-54817MEDIUM 6.5

  FluxBuilder's MStore API contains an authentication bypass vulnerability that can be exploited through alternate paths or channels to compromise password recovery functionality. An attacker can exploit this without authentication to modify data or disrupt service availability. The vulnerability affects MStore API versions up to and including 4.18.4.

• CVE-2026-54911MEDIUM 6.5

  UltraJSON, a widely-used C-based JSON library with Python bindings, contains a flaw in how it handles invalid UTF-8 sequences when the reject_bytes option is disabled. Rather than rejecting malformed data, the library silently converts these corrupted byte sequences into different Unicode characters, allowing attackers to bypass input validation checks and potentially manipulate data integrity. This vulnerability affects versions prior to 5.13.0 and requires developers to upgrade to patch this issue.

• CVE-2026-55078MEDIUM 6.5

  Coder, a platform for provisioning remote development environments through Terraform, contains a denial-of-service flaw in its file upload API. When authenticated users upload ZIP files, the system decompresses them into memory without enforcing a total size limit—only individual file limits. An attacker with legitimate upload credentials can craft a specially-prepared ZIP to consume all available memory, crashing the service. The flaw affects versions 2.17.0 through 2.34.1; patched versions 2.29.7, 2.32.7, 2.33.8, and 2.34.2 are now available.

• CVE-2026-55180MEDIUM 6.5

  pnpm, a widely-used Node.js package manager, has a credential leakage vulnerability in versions before 10.34.2 and 11.5.3. When a developer clones or installs a malicious repository, pnpm will expand environment variable placeholders (like ${API_KEY}) found in the repository's .npmrc or pnpm-workspace.yaml files. These expanded values—which may contain sensitive credentials or secrets—are then sent to an attacker-controlled package registry during dependency resolution, before any build scripts run. A developer with access to systems containing environment secrets is the attack vector; the attacker needs the victim to work with a malicious repository.

• CVE-2026-55197MEDIUM 6.5

  Hermes WebUI versions prior to 0.51.443 suffer from an access control flaw that lets authenticated users read conversation transcripts from other user profiles. An attacker with login credentials can craft requests to the /api/session endpoint specifying another user's session ID, bypassing the intended profile boundary protections. This exposure leaks sensitive conversation history and metadata that should remain private to each profile.

• CVE-2026-55198MEDIUM 6.5

  Hermes WebUI versions before 0.51.443 contain a flaw that allows logged-in users to download chat sessions belonging to other user profiles. The vulnerability exists in the session export feature, which fails to properly check that a user owns the session they're requesting. An attacker with valid login credentials can export transcripts from any session if they know or can guess its identifier, potentially exposing sensitive conversations, personal data, or proprietary information shared in those sessions.

• CVE-2026-55434MEDIUM 6.5

  Coder, a platform that manages remote development environments through Terraform, contains a memory exhaustion vulnerability in its AI Bridge feature. An authenticated user with access to AI Bridge endpoints can send unusually large requests that cause the application to consume excessive memory, rendering the service unavailable. The vulnerability affects versions 2.33.0 through 2.33.7 and 2.34.0 through 2.34.1. Patched versions 2.33.8 and 2.34.2 are available.

• CVE-2026-55490MEDIUM 6.5

  OpenWrt's Emergency Access Daemon contains a flaw that allows any attacker on your local network to crash it by sending a specially crafted network packet. The vulnerability exists in how the daemon handles message length calculations—it performs arithmetic that underflows (wraps around to an extremely large number), which is then misused as a copy size, leading to a denial of service. This affects OpenWrt versions before 25.12.5.

• CVE-2026-55514MEDIUM 6.5

  vLLM, a popular library used to deploy and run large language models, has a flaw that can crash the entire server when specific requests are sent to it. An authenticated user can send a specially crafted text completion request that causes the inference engine to fail an internal safety check and shut down the application. This affects vLLM versions 0.12.0 through 0.23.x when using models with M-RoPE (a rotary position embedding technique). The vulnerability was fixed in version 0.24.0.

• CVE-2026-55646MEDIUM 6.5

  vLLM versions 0.22.0 through 0.23.0 contain a denial-of-service vulnerability in their audio transcription and translation API endpoints. When a user uploads an audio file to these routes, the system reads the entire file into memory before checking whether it exceeds the configured size limit. An authenticated attacker can exploit this by uploading an extremely large audio file, forcing vLLM to consume massive amounts of memory during processing. Depending on how the service is deployed, this can cause severe memory pressure, trigger out-of-memory errors, or crash the vLLM process entirely. The vulnerability was patched in version 0.24.0.

• CVE-2026-55699MEDIUM 6.5

  pnpm, a widely-used Node.js package manager, contains a path-traversal vulnerability in how it handles global package installation and removal. Specifically, malicious package manifests can specify bin names (executable entry points) that bypass pnpm's validation checks—using names like ".", "..", or empty strings. When a user installs such a package globally and later removes, updates, or adds a replacement, pnpm incorrectly derives these malicious bin names and attempts to delete files at unintended locations. A bin name of "." would target the entire global bin directory; ".." would target its parent directory. An attacker distributing a malicious package could potentially delete or corrupt critical files in the global bin environment on a user's system.

• CVE-2026-55955MEDIUM 6.5

  Apache Tomcat contains an authentication weakness that allows attackers to replay encrypted cluster communications. When multiple Tomcat servers are configured to work together in a cluster, they use encryption to protect inter-server messages. This vulnerability lets an attacker capture and resend those encrypted messages without needing to know the encryption key, potentially gaining unauthorized access or making unauthorized changes. The flaw affects several versions of Tomcat across multiple release branches.

• CVE-2026-55956MEDIUM 6.5

  Apache Tomcat contains a flaw where security rules meant to restrict access to the default servlet don't properly enforce method-based restrictions. An attacker could potentially bypass these access controls by exploiting how the server handles HTTP method filtering, allowing unauthorized access to resources that should be protected. This affects multiple active versions of Tomcat across the 7.x, 8.5.x, 9.x, 10.1.x, and 11.x release lines.

• CVE-2026-55962MEDIUM 6.5

  A flaw in wolfSSL's TLS 1.3 implementation allows servers to incorrectly accept client authentication messages without verifying the client actually sent the required cryptographic proof of identity. This happens specifically when a server requests client authentication after the initial handshake is complete. The vulnerability only affects servers using wolfSSL's post-handshake authentication feature and does not impact typical clients or servers that perform all authentication during the initial connection setup.

• CVE-2026-56013MEDIUM 6.5

  A flaw in the License Manager for WooCommerce plugin (versions up to 3.0.15) allows unauthenticated users to directly access and modify license data they shouldn't have permission to view or change. The vulnerability stems from missing authorization checks on license-related endpoints, enabling attackers to tamper with license records without authentication. This could lead to license fraud, service disruption, or unauthorized feature access on affected e-commerce sites.

• CVE-2026-56024MEDIUM 6.5

  A cross-site request forgery (CSRF) vulnerability exists in WP EasyPay, a WordPress payment plugin by Saad Iqbal, affecting versions up to and including 4.5.0. CSRF flaws allow attackers to trick authenticated users into performing unintended actions—such as modifying settings, initiating transactions, or changing configuration—without their knowledge. Because this vulnerability requires no special privileges or authentication from the attacker's perspective, any unauthenticated user can craft a malicious link or embed it in a webpage to exploit WordPress site administrators.

• CVE-2026-56048MEDIUM 6.5

  A vulnerability in the Payment Gateway Based Fees and Discounts plugin for WooCommerce (versions 3.0.0 and earlier) allows attackers to modify order fees and discounts without authentication. An attacker could directly manipulate objects in the system by bypassing normal authorization checks, potentially altering pricing, applying unauthorized discounts, or modifying fees on orders they don't own. This is possible because the plugin fails to properly validate user permissions before allowing changes to fee and discount data.

• CVE-2026-56050MEDIUM 6.5

  A security flaw in Themeisle's PPOM (Product Personalization Options Manager) plugin for WooCommerce allows attackers to bypass access controls and make unauthorized changes to protected data. The vulnerability affects all versions up to and including 33.0.18. Attackers can exploit misconfigured permission settings without authentication to modify site content or cause service disruptions, though they cannot read sensitive information directly.

• CVE-2026-56077MEDIUM 6.5

  PraisonAI versions before 1.5.115 fail to enforce unique agent identifiers in their MultiAgentLedger system. An authenticated attacker can register multiple agents using the same ID, causing them to share the same ledger instance. This allows the attacker to view system prompts, conversation history, and other sensitive data intended for other agents—essentially eavesdropping on inter-agent communications and extracting configuration details that should remain isolated.

• CVE-2026-56079MEDIUM 6.5

  Capgo versions before 12.128.2 contain a critical authorization flaw that allows attackers with a read-only API key scoped to one organization to view webhook secrets and event logs belonging to other organizations. An attacker can extract HMAC signing keys and stored webhook payloads, then use this information to forge webhook events impersonating legitimate services. This breaks the tenant isolation that should prevent one customer's API credentials from accessing another customer's data.

• CVE-2026-56116MEDIUM 6.5

  dhcpcd, a widely-used DHCP client daemon on Unix-like systems, contains a memory leak flaw in how it processes IPv6 Router Advertisements. An attacker on the same network link can repeatedly send specially crafted Router Advertisements to exhaust the daemon's memory, eventually crashing it and disrupting network connectivity. The attacker needs no credentials and the traffic stays local to the link, making this a concern primarily for shared networks or systems where untrusted devices can reach the same link segment.

• CVE-2026-56148MEDIUM 6.5

  CVE-2026-56148 is a denial-of-service vulnerability in Elasticsearch that allows authenticated users to crash or disable individual cluster nodes by sending specially crafted queries. The flaw stems from uncontrolled recursion in query processing, which causes the server to consume excessive memory and CPU resources until it becomes unresponsive. An attacker must have valid credentials to exploit this, but once authenticated, no special permissions or user interaction is required—the malicious query does its damage simply by being processed.

• CVE-2026-56150MEDIUM 6.5

  Fleet Server, Elastic's agent management platform, contains a vulnerability in its file upload functionality that allows authenticated attackers to cause denial of service by submitting requests designed to consume excessive memory. An attacker with valid credentials can trigger uncontrolled resource allocation, potentially rendering the server unresponsive or forcing a restart. The vulnerability stems from missing limits or throttling mechanisms on resource consumption during upload operations.

• CVE-2026-56151MEDIUM 6.5

  A flaw in Kibana's Fleet policy input validation allows authenticated users to submit specially crafted data that crashes or disables Fleet functionality. An attacker with valid credentials can manipulate policy inputs to knock out agent communications, server operations, and policy management until administrators restart services or clear the offending configuration. This is a denial-of-service attack requiring legitimate access, so the blast radius is limited to environments where insider threats or compromised accounts are a concern.

• CVE-2026-56220MEDIUM 6.5

  Capgo before version 12.128.2 has an authorization bypass flaw that lets read-only team members modify over-the-air (OTA) update manifests. An attacker with read-only access to an organization can inject malicious manifest entries pointing to attacker-controlled S3 buckets. When devices check for updates via the public, unauthenticated endpoint, they receive poisoned metadata directing them to download malicious assets instead of legitimate ones. This bridges the gap between limited org access and the ability to compromise device software delivery.

• CVE-2026-56221MEDIUM 6.5

  Cap-go versions prior to 12.128.2 allow authenticated users with basic read-level API permissions to perform SQL injection attacks. By manipulating specific API parameters—including deviceIds, search, version_name, cursor, and actions—attackers can inject malicious SQL code to bypass access controls and view analytics data belonging to other users or applications. This is a logic-based flaw in how user input is incorporated into database queries rather than a result of missing authentication.

• CVE-2026-56229MEDIUM 6.5

  Capgo versions before 12.128.2 have a flaw that lets authenticated users view build logs and status information from applications they shouldn't have access to. An attacker with API credentials for one application can trick the system into revealing build details from a different application by mixing and matching app identifiers with job identifiers from unauthorized apps. This could expose sensitive build logs, metadata, and potentially hardcoded credentials.

• CVE-2026-56251MEDIUM 6.5

  Capgo versions prior to 12.128.2 contain a flaw in their access control system that lets already-authenticated administrators escalate themselves to a super_admin role without proper authorization. The vulnerability stems from broken row-level security (RLS) policies on the org_users table, meaning the database isn't correctly restricting which rows a user can modify. An attacker with admin credentials can exploit this to gain super_admin privileges, effectively taking full control of the system.

• CVE-2026-56262MEDIUM 6.5

  Crawl4AI, a web crawling and monitoring library, contains a security flaw that allows anyone on the network to perform damaging operations without logging in. Specifically, attackers can reach an administrative cleanup endpoint and disrupt the monitoring service's state, potentially causing outages or data loss. This affects versions before 0.8.7.

• CVE-2026-56273MEDIUM 6.5

  Flowise versions before 3.1.0 contain a security flaw that allows authenticated users with valid API tokens to write data to unintended locations on the server's filesystem. An attacker exploiting this vulnerability could direct vector store data (embeddings and related metadata) to arbitrary paths, potentially overwriting critical files or achieving code execution on the affected system. The vulnerability stems from insufficient validation of file path parameters, a common pattern known as path traversal. Because exploitation requires a valid API token, the risk is limited to threats from compromised credentials or malicious insiders with legitimate access.

• CVE-2026-56277MEDIUM 6.5

  Flowise versions before 3.1.2 contain a cross-origin request forgery vulnerability in their text-to-speech endpoint. The affected endpoint incorrectly permits any website to trigger TTS generation by hardcoding the CORS header to accept all origins (*), bypassing the application's otherwise secure default settings. An attacker can craft a malicious webpage that, when visited by a logged-in Flowise user, silently generates speech synthesis requests using that user's stored session credentials—a technique known as cross-origin credential abuse.

• CVE-2026-56302MEDIUM 6.5

  Capgo versions before 12.128.2 expose a critical cloud storage misconfiguration where app icons are stored without any access controls. This means anyone on the internet—without needing to log in or provide credentials—can view, upload, and delete these icons. Attackers can weaponize this to wipe out all stored icons and harvest sensitive identifiers like app IDs and user IDs that are embedded in or associated with those files.

• CVE-2026-56304MEDIUM 6.5

  Picklescan, a Python security tool, has a vulnerability in versions before 1.0.1 that allows attackers to craft malicious pickle payloads to create arbitrary zero-byte files on affected systems without authentication. While the vulnerability cannot directly execute code, it can be weaponized to create lock files or filesystem artifacts that disrupt application functionality. This attack is possible because the tool doesn't properly validate pickle data before deserializing it, a known dangerous practice in Python. An attacker only needs network access to send a crafted pickle payload to an application using a vulnerable version of picklescan.

• CVE-2026-56312MEDIUM 6.5

  Capgo versions prior to 12.128.2 contain a flaw in their account invitation system that allows attackers to create unauthorized accounts by bypassing the captcha protection mechanism. When a user submits an invitation acceptance request, the system creates the account before validating the captcha token, meaning an attacker can send requests with invalid or missing captcha data and still successfully register accounts. This vulnerability enables account takeover of invite links and account enumeration attacks.

• CVE-2026-56328MEDIUM 6.5

  Capgo, a mobile app update management platform, has a vulnerability that allows authorized users to create confusion about which app updates clients receive. The system normally supports multiple public update channels for the same application, but when clients request updates without specifying a channel preference, the system secretly picks one 'winner' channel. An authenticated app or channel manager can exploit this by creating overlapping channels, making it unclear which update bundle a requesting client will actually get. This breaks the predictability of release routing and could allow an attacker with insider access to stealthily serve unintended updates to users.

• CVE-2026-56335MEDIUM 6.5

  Capgo versions before 12.128.2 have a flaw that lets attackers with write-level API keys bypass security checks and directly modify protected channel settings in ways that should be prevented. An attacker could change sensitive flags like whether a channel is public or if it allows emulator connections, potentially exposing functionality or bypassing security controls that the application intended to enforce. The vulnerability exists because a database trigger that should prevent unauthorized changes has a null-check flaw that allows PostgREST requests with valid write credentials to slip through.

• CVE-2026-56346MEDIUM 6.5

  AVideo versions up to 25.0 contain a critical flaw in its message decryption endpoint that fails to require authentication. An attacker can send PGP-encrypted messages along with private keys and passphrases directly to the server, which will decrypt them without checking whether the requester has legitimate access. This exposes sensitive cryptographic material and creates opportunities for attackers to exhaust server resources by flooding the endpoint with decryption requests.

• CVE-2026-56394MEDIUM 6.5

  Craft CMS versions from 4.0.0-RC1 onward contain a weakness in how it handles file requests through its icon endpoint. An attacker who has valid login credentials can manipulate the file path parameter to read SVG files they shouldn't normally access on the server. The vulnerability stems from insufficient validation of the file extension before the system checks whether files exist, allowing an authenticated user to bypass intended access controls and retrieve sensitive local files.

• CVE-2026-56402MEDIUM 6.5

  NanoClaw versions before 2.1.17 contain a privilege escalation flaw that allows authenticated users to approve or reject sensitive actions—such as package installations—without having the authorization role to do so. An attacker who has legitimate access to the system can exploit a missing validation check in the approval response handler to perform privileged operations they shouldn't be able to execute, potentially installing malicious packages or disrupting system integrity.

• CVE-2026-56409MEDIUM 6.5

  xmlwf, a command-line XML parser tool included in libexpat, contains an integer overflow vulnerability when processing the output directory parameter (-d). When a user specifies a custom output directory for parsed XML files, the tool miscalculates buffer sizes for the output filename, potentially allowing an attacker to corrupt memory and execute arbitrary code or crash the application. This vulnerability requires local system access and user interaction to trigger.

• CVE-2026-56460MEDIUM 6.5

  HCL DevOps Deploy and HCL Launch contain an information disclosure vulnerability that allows authenticated users to retrieve sensitive configuration data and secrets through API responses. An attacker with valid credentials could leverage this exposure to understand system architecture, extract credentials, and plan follow-up attacks. The vulnerability requires prior authentication, which moderates immediate risk but significantly impacts organizations relying on these platforms for infrastructure automation.