CVE Explorer

pageant provides a PageantStream type that implements AsyncRead and AsyncWrite traits and can be used to communicate with a running Pageant instance. From 0.2.0 until 0.2.5, PageantStream::new in pageant/src/wmmessage.rs trusts a peer-controlled four-byte length prefix and buffers until len + 4 bytes are available without checking the fixed _AGENT_MAX_MSGLEN transport capacity. An application that bridges an SSH forwarded-agent channel to the raw stream can allow a malicious server to force large allocations or retain a partially filled frame, causing client memory exhaustion. This issue is fixed in version 0.2.5.

russh is a Rust SSH client and server library that includes an SSH agent protocol implementation. From 0.60.0 until 0.65.0, channel-open parsing accepts a peer-declared maximum_packet_size of zero and passes it into ChannelTx::poll_writable in russh/src/channels/io/tx.rs. The Channel::data, Channel::make_writer, and Channel::into_stream asynchronous write paths then calculate a writable size of zero and remain pending forever because window adjustments cannot satisfy a static zero packet maximum. An authenticated client can strand server writes, and a malicious server can strand client writes, allowing affected tasks, channels, and connections to accumulate without surfacing an error. This issue is fixed in version 0.65.0.

russh is a Rust SSH client and server library that includes an SSH agent protocol implementation. Prior to 0.65.0, the client Session::process_packet path in russh/src/client/encrypted.rs accepts repeated packets encoded with russh's private msg::EXT_INFO = 7 before authentication, while the standards-compliant EXT_INFO message number 82 does not reach the affected handler. The handle_ext_info function appends every parsed extension name to received_extensions without clearing, deduplicating, or bounding the vector. A malicious SSH server that deliberately sends private message number 7 can therefore retain an unbounded number of attacker-controlled extension names for the session lifetime, causing progressive client memory exhaustion. This issue is fixed in version 0.65.0.

russh is a Rust SSH client and server library that includes an SSH agent protocol implementation. Prior to 0.65.0, russh::keys::agent::server::serve in russh/src/keys/agent/server.rs accepts repeated SSH_AGENT_CONSTRAIN_CONFIRM records while adding a key and appends every Constraint::Confirm record to the key's retained constraint vector without a count or memory limit. A process that can connect to the agent socket, including the remote side of a forwarded agent session, can repeatedly add constrained keys and cause renewable memory growth that persists for the process lifetime because keys without a lifetime constraint do not expire. This issue is fixed in version 0.65.0.

Crass is a Ruby CSS parser that is compliant with the CSS Syntax Level 3 specification. Prior to 1.0.8, Crass incorrectly treats a closing parenthesis after an escaped backslash as escaped while consuming the remnants of a bad URL. Crass therefore keeps declarations, style rules, or at-rules inside a bad URL token even though a browser treats them as live CSS. A sanitizer that parses untrusted CSS with Crass and reserializes it without rejecting the bad URL can allow CSS used for UI redress, external resource loading, or CSS-based data exfiltration, but this issue does not enable script execution in modern browsers. This issue is fixed in version 1.0.8.

OpenTelemetry Python provides the Python API and SDK for OpenTelemetry. Prior to 1.45.0, JaegerPropagator.extract() in opentelemetry-propagator-jaeger processes inbound uberctx-* baggage values without entry, per-entry byte, or total-byte limits. Services that enable the optional Jaeger propagator can perform unnecessary CPU work and short-lived heap allocations when an untrusted carrier supplies numerous or oversized values, contributing to denial of service when transport-level limits do not constrain the input. Common default Python HTTP server limits reduce the practical impact, while custom getters, non-HTTP transports, or raised header limits increase exposure. This issue is fixed in version 1.45.0.

dd-trace is the Datadog APM client for Node.js. From 2.0.0 until 5.131.1 and 6.20.1, the default DD_TRACE_OBFUSCATION_QUERY_STRING_REGEXP can perform excessive backtracking on attacker-controlled URL query strings. The expression is applied synchronously while recording HTTP URL metadata, allowing a crafted request to consume excessive CPU, stall request processing, and make the instrumented application unavailable. This issue is fixed in versions 5.131.1 and 6.20.1.

MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. From 3.12.0rc0 until 3.17.0, third-party scorer deserialization in mlflow/genai/scorers/base.py accepts request-selected descendants of THIRD_PARTY_SCORER_ALLOWED_MODULES rather than requiring an exact allowed module. A caller can use MLflow artifact endpoints to place a Python module beneath a writable allowed package and submit its module and class through /ajax-api/3.0/mlflow/scorer/invoke. Scorer.model_validate_json() imports the module and constructs the class before trace processing. Authentication-disabled deployments permit this code execution without credentials, while basic-auth deployments require a principal capable of creating an experiment and run and uploading artifacts. In either configuration, the allowed scorer package must be writable by the server process. This issue is fixed in version 3.17.0.

MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. From 2.19.0 until 3.17.0, mlflow.crewai.autolog() records CrewAI Agent objects without consistently sanitizing nested credential-bearing values. mlflow/crewai/autolog.py::_get_agent_attributes() converts the Agent llm field to a string that includes the provider API key, while mlflow/crewai/autolog.py::_construct_full_inputs() expands the Agent object into mlflow.spanInputs. Applications can therefore send plaintext provider credentials to the MLflow trace backend or a configured OTLP destination, where trace readers can obtain them. This issue is fixed in version 3.17.0.

MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. From 3.10.0 until 3.17.0, the prompt branch in _create_model_version() accepts schemeless relative sources because path validation runs only when a source has a URL scheme, and mlflow.prompt.is_prompt is interpreted by tag presence rather than a true value. A remote caller can store a server-local path as the model version artifact location and use /api/2.0/mlflow/model-versions/get-artifact to read files accessible to the MLflow server process. Default unauthenticated servers require no privileges, while basic-auth deployments are exposed to authenticated users who can create registered models. This issue is fixed in version 3.17.0.

follow-redirects is a Node.js HTTP and HTTPS client wrapper that automatically follows redirects. From 1.16.0 until 1.16.2, the documented sensitiveHeaders option is checked with value instanceof Array, so an array created in another JavaScript realm is replaced with an empty list even though Array.isArray recognizes it. When a request then follows an attacker-controlled cross-origin redirect, a custom credential header named in that discarded list can be forwarded to the redirect destination. This issue is fixed in version 1.16.2.

CKEditor 5 is a modern JavaScript rich-text editor with an MVC architecture. From 45.0.0 until 47.7.4 and 48.5.1, the engine uses a vulnerable es-toolkit version whose prototype-pollution behavior can be reached while CKEditor processes incoming style attribute values. Crafted editor content can thereby cause unauthorized JavaScript execution in the browser. This issue is fixed in versions 47.7.4 and 48.5.1.

CKEditor 5 is a modern JavaScript rich-text editor with an MVC architecture. From 29.0.0 until 47.7.4 and 48.5.1, specially crafted markup can trigger cross-site scripting when General HTML Support is enabled and specifically configured to allow object content. Processing that content can execute attacker-controlled JavaScript in a browser context isolated from the origin of the application embedding the editor. This issue is fixed in versions 47.7.4 and 48.5.1.

The Model Context Protocol Python SDK is the official Python SDK for Model Context Protocol servers and clients. From 1.10.0 until 1.30.0 and 2.2.0, ClientSession, ClientSessionGroup, and Client tool calls validate server-declared outputSchema documents with jsonschema's default resolver, causing an unresolved $ref value to fetch an HTTP, HTTPS, or file URL from the client's network and filesystem context. A server can trigger blind requests or local-file reads and can stall every session in the process because resolution runs synchronously on the event loop without a timeout. This issue is fixed in versions 1.30.0 and 2.2.0.

The Model Context Protocol Python SDK is the official Python SDK for Model Context Protocol servers and clients. From 1.8.0 until 1.30.0 and 2.2.0, streamable_http_client, streamablehttp_client, sse_client, ClientSessionGroup, OAuth, and create_mcp_http_client client paths enable follow_redirects and follow redirects to a different origin, forwarding caller-supplied custom credential headers and, for 307 or 308 responses, the request body. An attacker who can influence a trusted endpoint's redirect destination can receive that material. This issue is fixed in versions 1.30.0 and 2.2.0.

The Model Context Protocol Python SDK is the official Python SDK for Model Context Protocol servers and clients. From 1.10.0 until 1.30.0 and 2.2.0, server-side bearer authentication configured through AuthSettings with token_verifier accepts a valid token issued for another service when the authorization server serves multiple resources and the TokenVerifier does not independently validate the aud audience. This resource-confusion flaw lets a token holder authenticate to the wrong MCP server and exercise that token's resulting access. This issue is fixed in versions 1.30.0 and 2.2.0.

The MembershipWorks plugin for WordPress before 6.17 contains an information disclosure vulnerability that allows unauthenticated attackers to access member-only content through the WordPress REST API. Because the filter that strips protected content for non-members is only attached during normal front-end request handling and not during REST API requests, an unauthenticated attacker can retrieve the raw member-only content in a post or page’s rendered content, bypassing the plugin’s content protection.

Sunhillo SureLine before 9.1.5 contains an authentication bypass vulnerability that allows an unauthenticated remote attacker to impersonate any user with an active web session by forging a session cookie containing only a cleartext username with no random identifier or integrity protection. Attackers can guess default account names such as Admin or sftp_user and submit a forged username cookie to satisfy the login check in checkLogin (libwebcmn.so) and networkDiag.cgi, chaining into root command execution.

Sunhillo SureLine before 9.1.5 contains an OS command injection vulnerability in the networkDiag.cgi endpoint that allows an authenticated remote attacker with any web account to execute arbitrary commands as root by injecting shell metacharacters into the ipAddr or dnsAddr parameters. Attackers can bypass the incomplete character denylist from the prior CVE-2021-36380 fix using URL-encoded newline (%0a) or ampersand (%26) sequences that act as command separators after decoding in the ping, nslookup, or traceroute diagnostic modes, leading to full system compromise.

Sunhillo SureLine before 9.1.5 contains a path traversal vulnerability in the radarUpdater.cgi endpoint that allows unauthenticated remote attackers to read arbitrary files with root privileges due to missing authentication checks and insufficient sanitization of the filename parameter. Attackers can use directory traversal sequences to access sensitive files such as /etc/shadow, the web user database, the root SSH private key, and the device's TLS private key, leading to full system compromise.

Response body and chunk parsing would handle non-hex or negative integers incorrectly which could result in misinterpreting HTTP response contents.

When logging in to a Matrix account, Thunderbird used the homeserver address provided in the server's login response without validating it. A malicious or compromised homeserver could provide a file: URL, causing Thunderbird to send all further Matrix requests of that session to local files or, on Windows, to an attacker-controlled network share, which could leak the user's NTLM credentials. The NTLM credential leak only affects Thunderbird on Windows. Other operating systems are unaffected. This vulnerability was fixed in Thunderbird 159 and Thunderbird 153.6.

Clicking a news: or snews: link that referenced a message-ID made Thunderbird connect, without asking, to the host and port given in the link, even if it was not a configured news server. Line breaks in the message-ID were sent to the server unmodified, so a crafted link could send arbitrary commands to services reachable from the user's network. In addition, snews: links to such servers connected without TLS. This vulnerability was fixed in Thunderbird 159 and Thunderbird 153.6.

A remote RSS feed could use the itunes:new-feed-url or newLocation element to permanently change its stored subscription URL to a file: URL. Thunderbird then read that location on every feed update. This allowed a malicious feed, or an attacker able to modify a feed delivered over HTTP, to make Thunderbird read local files and, on Windows, connect to an attacker-controlled network share, which could leak the user's NTLM credentials. The NTLM credential leak only affects Thunderbird on Windows. Other operating systems are affected by the local file read only. This vulnerability was fixed in Thunderbird 159 and Thunderbird 153.6.

When looking for calendars in a directory on a server, during account setup or when adding a new network calendar, Thunderbird accepted calendar locations pointing to other servers, local files or network shares. A malicious server, or a network attacker when the lookup used plain HTTP, could make Thunderbird subscribe to such a location. On Windows this could leak the user's NTLM credentials to an attacker-controlled host on every calendar refresh, and changes to the calendar could be written to the attacker's share or overwrite a local file. This vulnerability was fixed in Thunderbird 159 and Thunderbird 153.6.