CVE Feed

Improper Neutralization of CRLF Sequences ('CRLF Injection') vulnerability in ninenines cowlib allows HTTP request splitting and cookie smuggling via unvalidated cookie name and value fields. cow_cookie:cookie/1 in cowlib builds a client-side Cookie: request header from a list of name-value pairs without validating either field. An attacker who controls the cookie names or values passed to this function can inject ;, ,, CR, LF, or TAB characters into the serialized header. This enables two classes of attack: cookie smuggling within a single header (e.g. injecting ; admin=1 to introduce a phantom cookie that the receiving server treats as authentic) and HTTP request header splitting (injecting CRLF to append arbitrary headers or smuggle a complete second request against a shared upstream proxy). The decoder side (parse_cookie_name/1, parse_cookie_value/1) and setcookie/3 already validate and reject these characters; the encoder alone is missing the check. This issue affects cowlib: from 2.9.0 onward.

Summarize versions through 0.14.1, fixed in commit 0cfb0fb, creates the daemon configuration directory and file with default filesystem permissions that may be world-readable on Unix-like systems, allowing local attackers to read bearer tokens and API credentials stored in ~/.summarize/daemon.json. A local attacker can exploit these permissive permissions to read the daemon bearer token and persisted provider credentials, enabling unauthorized access to the daemon or recovery of sensitive API keys.

A security flaw has been discovered in VectifyAI PageIndex up to f50e52975313c6716c02b20a119577a1929decba. Affected by this vulnerability is the function toc_transformer of the file pageindex/page_index.py of the component PDF Table of Contents Handler. The manipulation results in infinite loop. The attack may be launched remotely. The exploit has been released to the public and may be used for attacks. This product operates on a rolling release basis, ensuring continuous delivery. Consequently, there are no version details for either affected or updated releases.

Inbox Zero is an AI personal assistant for email. Prior to 2.29.3, the cleaner email stream endpoint used a shared Redis subscription listener, which could deliver thread events for one authenticated account to another authenticated account using the cleaner feature at the same time. This vulnerability is fixed in 2.29.3.

Flowise is a drag & drop user interface to build a customized large language model flow. Prior to 3.1.0, multiple tool implementations directly import and invoke raw HTTP clients (node-fetch, axios) instead of using the secured wrapper. These tools include (1) OpenAPIToolkit/OpenAPIToolkit.ts, (2) WebScraperTool/WebScraperTool.ts, (3) MCP/core.ts, and (4) Arxiv/core.ts. This vulnerability is fixed in 3.1.0.

Network-AI is a TypeScript/Node.js multi-agent orchestrator. Prior to 5.1.3, the MCP HTTP transport accepts JSON-RPC tools/call requests with no authentication, session, origin, or token check, and dispatches them directly to the orchestrator's tool registry. The default bind address is 0.0.0.0. As a result, any party with network reachability to the service can enumerate and invoke privileged management tools. This vulnerability is fixed in 5.1.3.

In JetBrains TeamCity before 2026.1 2025.11.5 authenticated users could expose server API to unauthorised access

The Open edx Enterprise Service app provides enterprise features to the Open edX platform. From 7.0.2 to 7.0.4, the sync_provider_data endpoint in SAMLProviderDataViewSet fetches SAML metadata from a URL stored in SAMLProviderConfig.metadata_source. An authenticated user with the Enterprise Admin role can set this field to an arbitrary URL via the SAMLProviderConfigViewSet PATCH endpoint, then trigger a server-side HTTP request by calling sync_provider_data. The fetch in fetch_metadata_xml() passes the URL directly to requests.get() with no scheme enforcement, IP filtering, or timeout. This vulnerability is fixed in 7.0.5.

Neat VNC is a VNC server library. Prior to 0.9.6, a pre-authentication stack buffer overflow exists in neatvnc in the RSA-AES security type handler. An unauthenticated remote attacker who can reach the VNC listening socket can send a crafted security type 5 (RSA-AES) or security type 129 (RSA-AES-256) handshake with an oversized client RSA public key, causing rsa_aes_send_challenge in src/auth/rsa-aes.c to overflow a 1024-byte on-stack buffer when encrypting the server challenge. This results in at least a denial of service via server crash. This vulnerability is fixed in 0.9.6.

Open edX Platform enables the authoring and delivery of online learning at any scale. The HTML sanitizer clean_thread_html_body() used for discussion notification emails fails to remove <style> tags from user-generated discussion post content. This content is rendered with Django's |safe template filter in email notification templates, allowing any enrolled student to inject arbitrary CSS into email notifications sent to other users. This enables email tracking (IP address disclosure), content spoofing, and phishing attacks. This vulnerability is fixed with commit cddc25cd791bb78f76833896e4778f668861df12.

Open edX Platform enables the authoring and delivery of online learning at any scale. The sync_provider_data endpoint in SAMLProviderDataViewSet allows authenticated Enterprise Admin users to supply an arbitrary URL via the metadata_url POST parameter. This URL is passed directly to requests.get() in fetch_metadata_xml() without any URL validation, IP filtering, or scheme enforcement. An attacker with Enterprise Admin privileges can force the server to make HTTP requests to internal network services, cloud metadata endpoints (e.g., AWS 169.254.169.254), or other attacker-controlled destinations. This vulnerability is fixed by commit 6fda1f120ff5a590d120ae1180185525f399c6d0 and 70a56246dd9c9df57c596e64bdd8a11b1d9da054.

jq is a command-line JSON processor. In 1.8.1 and earlier, unbounded recursion in jv_object_merge_recursive() allows a crafted jq program to crash the process with a segfault. The function is reachable through the * operator when both operands are objects.

jq is a command-line JSON processor. In 1.8.1 and earlier, jq accepts embedded NUL bytes in import paths at the jq-language level, but later resolves those paths through C string operations during module and data-file lookup. This creates a mismatch between the logical import string that policy or audit code may validate and the on-disk path that jq actually opens.

jq is a command-line JSON processor. In 1.8.2rc1 and earlier, the ordinary module loader recurses without cycle detection when two otherwise valid modules include each other.

jq is a command-line JSON processor. In 1.8.1 and earlier, when decNumberFromString is given a number literal of INT_MAX-1 (2147483646) digits, the D2U() macro overflows during signed-int arithmetic. The wrapped negative value bypasses the heap-allocation size check, causes the function to use a 30-byte stack buffer, and then writes ≈715 million 16-bit units (≈1.4 GiB) at an offset 1.43 GiB below the stack frame. The written content is fully attacker-controlled (the parsed decimal digits, packed 3-per-unit).

`xml.parsers.expat` and `xml.etree.ElementTree` use insufficient entropy for Expat hash-flooding protection, which allows a crafted XML document to trigger hash flooding.\r\n\r\nFully mitigating this vulnerability requires both updating libexpat to 2.8.0 or later and applying this patch.

jq is a command-line JSON processor. In 1.8.1 and earlier, Top-level jq programs loaded from a file with -f are truncated at the first embedded NUL byte on current upstream HEAD. A crafted filter file such as . followed by \x00 and arbitrary suffix compiles and executes as only the prefix before the NUL. This leaves jq with a post-CVE-2026-33948 prefix/full-buffer mismatch on the compilation path even though the JSON parser path has already been fixed.

An authenticated user with upload permission to a hosted repository can store content that causes arbitrary JavaScript to execute in the browser of any user who browses that repository directory via the HTML index page in Sonatype Nexus Repository versions 3.6.0 through versions before 3.92.0. This could allow the attacker to perform actions in the context of the victim's session.

jq is a command-line JSON processor. In 1.8.1 and earlier, jv_contains recurses into nested arrays/objects with no depth limit. With a sufficiently nested input structure (built programmatically with reduce, since the JSON parser caps at depth 10000), the C stack is exhausted.

Bitwarden Server prior to v2026.4.1 does not require master-password re-authentication when retrieving or rotating an organization's SCIM API key, allowing an authenticated user with SCIM management privileges to obtain the key using only a valid session.

jq is a command-line JSON processor. In 1.8.1 and earlier, the jq bytecode VM's data stack tracks its allocation size in a signed int. When the stack grows beyond ≈1 GiB (via deeply nested generator forks), the doubling arithmetic overflows. The wrapped value is passed to realloc and then used for a memmove with attacker-influenced offsets.

Bitwarden Server prior to v2026.4.0 contains a missing authorization vulnerability that allows a provider service user to add an arbitrary organization to their provider via `POST /providers/{providerId}/clients/existing`, resulting in takeover of the target organization; self-hosted installations are unaffected as this endpoint is restricted to Cloud via SelfHosted(NotSelfHostedOnly = true).

Bitwarden Server prior to v2026.4.1 contains a missing authorization vulnerability that allows any authenticated user to write ciphers into an arbitrary organization via `POST /ciphers/import-organization` by submitting an empty `collections` array, which causes the server-side permission check to be skipped.

An authenticated administrator who configures or tests LDAP connectivity in Sonatype Nexus Repository Manager versions 3.0.0 through 3.91.1 may be able to initiate unintended server-side connections when interacting with a malicious LDAP server.

Zen is a firefox-based browser. Prior to 1.19.12b, the ZEN Browser incorrectly truncates long hostnames in the address bar and shows only the attacker-controlled prefix of the subdomain, hiding the actual registrable domain (eTLD+1). As a result, an attacker can craft extremely long malicious subdomains that visually imitate trusted brands, and the browser will display only the spoofed prefix, misleading users about the actual origin of the site. This directly compromises the URL bar as a security indicator and creates a phishing/supply-chain attack vector. This vulnerability is fixed in 1.19.12b.

Zen is a firefox-based browser. Prior to 1.19.12b, RSS feed URLs entered by the user are validated to http: or https: in promptForFeedUrl, but item links inside the feed are not subject to the same restriction. The provider maps each RSS/Atom item link into item.url, filters only for presence and date, and returns the item list. The live-folder manager later creates pinned lazy tabs from these values with gBrowser.addTrustedTab(item.url, ...). This vulnerability is fixed in 1.19.12b.

** REJECT ** DO NOT USE THIS CANDIDATE NUMBER. Reason: This candidate was issued in error. Notes: All references and descriptions in this candidate have been removed to prevent accidental usage.

Exposure of Sensitive Information to an Unauthorized Actor vulnerability in Wikimedia Foundation Echo. This vulnerability is associated with program files includes/Api/ApiEchoNotifications.Php. This issue affects Echo: from * before 1.43.7, 1.44.4, 1.45.2.

Zen is a firefox-based browser. Prior to 1.19.9b, Zen Browser ships a Mozilla Application Resource (MAR) updater (org.mozilla.updater) that has had all MAR signature verification stripped from the Firefox codebase it was forked from. The MAR files served to users contain zero cryptographic signatures, and the updater binary contains zero cryptographic verification code. This eliminates the defense-in-depth that MAR signing provides. If the update server or GitHub release pipeline is compromised, arbitrary unsigned code can be delivered to all Zen users via the auto-update mechanism. This vulnerability is fixed in 1.19.9b.

Vulnerability in Wikimedia Foundation MediaWiki. This vulnerability is associated with program files includes/Actions/ActionEntryPoint.Php, includes/Request/FauxResponse.Php. This issue affects MediaWiki: from * before 1.43.7, 1.44.4, 1.45.2.

Vulnerability in Wikimedia Foundation MediaWiki. This vulnerability is associated with program files includes/Page/Article.Php. This issue affects MediaWiki: from * before 1.43.7, 1.44.4, 1.45.2.

Taiga is a project management platform for startups and agile developers. Prior 6.9.1, Taiga front is vulnerable to stored XSS. This vulnerability is fixed in 6.9.1.

Exposure of Sensitive Information to an Unauthorized Actor vulnerability in Wikimedia Foundation MediaWiki. This vulnerability is associated with program files includes/Specials/SpecialUserRights.Php. This issue affects MediaWiki: from * before 1.43.7, 1.44.4, 1.45.2.

OpenClaw before 2026.4.23 contains an improper access control vulnerability in the gateway tool's config.apply and config.patch operations that allows compromised models to write unsafe configuration changes by bypassing an incomplete denylist protection. Attackers can persist malicious config modifications affecting command execution, network behavior, credentials, and operator policies that survive restart.

OpenClaw before 2026.4.23 caches resolved webhook route secrets backed by SecretRef values, allowing stale secrets to remain valid after rotation and reload. Attackers with previously valid webhook route secrets can continue authenticating requests and invoking configured webhook task flows until gateway or plugin restart.

OpenClaw before 2026.4.23 contains an arbitrary code execution vulnerability in the bundled plugin setup resolver that loads setup-api.js from process.cwd() during provider setup metadata resolution. Attackers can execute arbitrary JavaScript under the current user account by placing a malicious extensions/<plugin>/setup-api.js file in a repository and convincing a user to run OpenClaw commands from that directory.

OpenClaw before 2026.4.22 allows workspace dotenv files to override connector endpoint hosts for Matrix, Mattermost, IRC, and Synology connectors. Attackers with workspace access can redirect runtime traffic to malicious endpoints by setting endpoint variables in dotenv files.

OpenClaw before 2026.4.20 contains a hook session-key bypass vulnerability that allows attackers to circumvent the hooks.allowRequestSessionKey opt-in restriction. Attackers can render externally influenced session keys through templated hook mappings to bypass webhook routing isolation controls.

OpenClaw before 2026.4.20 contains a guard bypass vulnerability in the agent-facing gateway config.patch and config.apply endpoints that fails to protect operator-trusted settings including sandbox policy, plugin enablement, gateway auth/TLS, hook routing, MCP server configuration, SSRF policy, and filesystem hardening. A prompt-injected model with access to the owner-only gateway tool can persist unauthorized changes to protected operator settings.

OpenClaw before 2026.4.20 contains a server-side request forgery vulnerability in browser CDP profile creation that skips strict-mode SSRF policy checks. Attackers can create stored profiles pointing to private-network or metadata endpoints that bypass security policies and are later probed during normal profile status operations.

OpenClaw before 2026.4.20 fails to properly preserve untrusted labels for isolated cron awareness events, allowing webhook-triggered cron agent output to be recorded as trusted system events. Attackers can exploit this trust-labeling issue to strengthen prompt-injection attacks by rendering untrusted events as trusted System events.

OpenClaw before 2026.4.20 contains a tool policy bypass vulnerability allowing bundled MCP and LSP tools to circumvent configured tool restrictions. Attackers with local agent access can append restricted tools to the effective tool set after policy filtering, bypassing profile policies, allow/deny lists, owner-only restrictions, sandbox policies, and subagent policies.

OpenClaw before 2026.4.22 contains a security envelope constraint bypass vulnerability allowing restricted subagents to spawn ACP child sessions that fail to inherit depth, child-count limits, control scope, or target-agent restrictions. Attackers can exploit this by spawning child sessions that bypass subagent-only constraints, potentially escalating privileges or accessing restricted resources.

OpenClaw before 2026.4.15 contains an arbitrary local file read vulnerability in the webchat audio embedding helper that fails to apply local media root containment checks. Attackers can influence agent or tool-produced ReplyPayload.mediaUrl parameters to resolve absolute local paths or file URLs, read audio-like files, and embed them base64-encoded into webchat responses.

OpenClaw before 2026.4.20 contains an improper environment variable validation vulnerability in MCP stdio server configuration that allows attackers to execute arbitrary code. Malicious workspace configurations can pass dangerous startup variables like NODE_OPTIONS, LD_PRELOAD, or BASH_ENV to spawned MCP server processes, enabling code injection when operators start sessions using those servers.

OpenClaw before 2026.4.22 contains an authentication bypass vulnerability in the Control UI bootstrap config endpoint that allows unauthenticated attackers to read sensitive configuration fields. Attackers can access the bootstrap config route without a valid Gateway token to expose sensitive bootstrap and config information intended only for authenticated Control UI sessions.

OpenClaw before 2026.4.20 contains a message classification vulnerability in Feishu card-action callbacks that misclassifies direct messages as group conversations. Attackers can bypass dmPolicy enforcement by triggering card-action flows in direct message conversations that should have been blocked by restrictive policies.

OpenClaw versions 2026.4.5 before 2026.4.20 contain an environment variable injection vulnerability allowing workspace dotenv to override MINIMAX_API_HOST. Attackers can redirect credentialed MiniMax API requests to attacker-controlled origins, exposing the MiniMax API key in Authorization headers.

OpenClaw before 2026.4.21 contains an authorization bypass vulnerability in command-auth.ts that allows non-owner senders to execute owner-enforced slash commands when wildcard inbound senders are configured without explicit owner allowFrom settings. Attackers can exploit this by sending commands like /send, /config, or /debug on affected channels to bypass owner-only command authorization checks.

kafka-sink-azure-kusto Kafka Connect plugin is the official Microsoft sink for Azure Data Explorer (Kusto). Prior to 5.2.3, kafka-sink-azure-kusto did not sanitize user-controlled values inside the kusto.tables.topics.mapping configuration. The db, table, mapping, and format fields of each mapping entry were interpolated directly into KQL management/query commands via String.formatted(...) (e.g., FETCH_TABLE_COMMAND.formatted(table) → "<table> | count", FETCH_TABLE_MAPPING_COMMAND.formatted(table, format, mapping) → ".show table <table> ingestion <format> mapping '<mapping>'"). An actor able to influence the connector configuration (for example, someone with permissions to submit or edit Kafka Connect connector configs) could embed KQL metacharacters (;, |, ') to execute arbitrary management commands in the context of the connector's service principal — enabling schema enumeration/modification, ingestion-mapping tampering, or changes to streaming/retention policies on the target Azure Data Explorer database. This is a tampering vulnerability. Exploitation requires privileged access to the connector configuration; no end-user interaction or Kafka record payload is involved. This vulnerability is fixed in 5.2.3.