The russh SSH agent server does not call the Agent::confirm method before signing with keys added with the confirm constraint (ssh-add -c). This confirm constraint is intended to require user approval before signing. The issue was introduced in version 0.50.0-beta.12 and affects all versions up to and including 0.64.1. As a result, keys added with the confirm constraint are signed without user confirmation, allowing any process with access to the agent to use those keys without approval. This undermines the security purpose of the confirm constraint defined in RFC 9987.
The pageant crate's PageantStream component improperly handles a 4-byte length prefix from its input stream without validating the claimed frame size. This allows a malicious peer to claim an oversized frame (up to ~4 GiB) and cause unbounded memory buffering, leading to a remote memory-exhaustion denial-of-service (DoS) on the client process. The vulnerability affects versions from 0.2.0 up to but not including 0.2.5. A fix is available in version 0.2.5 that rejects frames exceeding the transport's maximum size before buffering.
A vulnerability in russh versions 0.60.0 up to but not including 0.65 allows a peer to advertise a maximum packet size of zero in channel-open messages. This causes the AsyncWrite path to stall indefinitely because the write operation never completes or errors, leading to a silent hang of the application task. The issue arises from lack of validation of the peer-declared maximum packet size and inconsistent handling in the AsyncWrite code path. This can be exploited by an authenticated client against a server or by a malicious server against a client to cause a denial of service by hanging write operations.
The russh SSH agent server (version prior to 0.65.0) has a vulnerability where it unboundedly allocates memory when processing SSH_AGENT_CONSTRAIN_CONFIRM (0x02) records during key addition requests. Each 1-byte confirm record causes a 56-byte allocation retained for the process lifetime, with no limit on the number of such records per key or the number of keys stored. This can lead to persistent memory growth controlled by any local process able to connect to the agent socket. The issue was fixed in version 0.65.0.
The russh SSH client versions prior to 0.65 have a vulnerability in their handling of EXT_INFO packets during the pre-authentication phase. The client accumulates extension names from these packets into a vector without any limit or deduplication, allowing a malicious server to cause unbounded memory growth by sending many such packets. This can lead to denial of service via resource exhaustion. The issue is fixed in version 0.65.
GraphQL-core versions from 3.3.0a3 up to but not including 3.3.1 contain a denial of service vulnerability caused by a memory leak during aborted validations. When a validation is aborted due to exceeding the max_errors limit, a shared ValidationAbortedError instance accumulates traceback frames, retaining memory indefinitely. This allows unauthenticated clients to cause memory exhaustion by sending documents with over 100 validation errors, potentially leading to out-of-memory kills. The issue is fixed in version 3.3.1 by creating a new error instance per aborted validation and clearing tracebacks.
OAuthlib versions prior to 4.0.0 contain a vulnerability in token grant handlers where an authenticated low-privilege client can supply a different client_id in the token request body. This causes grant type and scope validation to be performed under the attacker-supplied client_id rather than the authenticated client's identity. The flaw affects the client_credentials and resource owner password credentials grant flows. It allows a low-privilege client to obtain tokens with broader scopes or grant types assigned to another client. The issue is fixed in version 4.0.0.
OpenWISP Controller contains an authorization flaw in its authenticated REST API endpoint for downloading CA Certificate Revocation Lists (CRLs). The endpoint does not properly scope requests to the managed organization, allowing an organization manager to access CRLs belonging to other organizations by enumerating CA IDs. The exposure is limited to CRL data, which includes CA issuer information and revoked certificate serial numbers, but does not include private keys or allow modification. The flaw affects versions from 1.0.0 up to but not including 1.3.2 and has been fixed in version 1.3.2.
A vulnerability in Fiber v2.52.1 and v2.52.2 CORS middleware allowed wildcard origins to be matched incorrectly without enforcing a dot boundary or scheme check. This caused origins with hosts ending in the base domain, but not actual subdomains (e.g., https://evil-example.com for https://*.example.com), to be reflected in Access-Control-Allow-Origin with credentials allowed. The issue was fixed in v2.52.3 by improving the origin matching logic.
signalk-server versions from 2.28.0-beta.2 up to but not including 2.34.0 contain a server-side request forgery (SSRF) vulnerability via the app store icon fetcher. The server fetches absolute URLs declared in a plugin's signalk.appIcon field without proper restrictions, allowing an attacker to cause the server to make HTTP requests to internal or cloud metadata endpoints. Responses with image content types are stored and served back to the admin UI. This can be exploited by publishing a malicious plugin to the app store. A fix is available in version 2.34.0.
A path traversal vulnerability in the Signal K server's resources API allows unauthenticated or low-privileged users to read, write, and delete files outside the intended storage directory. This includes reading the security.json file containing the JWT signing secret, enabling attackers to forge admin JWT tokens and gain administrative access. With admin privileges, attackers can create providers that execute arbitrary shell commands, resulting in remote code execution with server privileges. The vulnerability affects all signalk-server versions prior to 2.34.0 and resources-provider versions prior to 1.5.3. The issue is patched in versions 2.34.0 and 1.5.3.
An integer overflow vulnerability exists in the ICO loader of imgproxy before version 4.0.16. The flaw occurs when a 32-bit attacker-controlled data size is added to a constant header length, causing an allocation size wraparound and resulting in a heap overflow. This can lead to denial of service or potentially remote code execution when processing a crafted ICO image.
A heap out-of-bounds write vulnerability exists in imgproxy's ICO loader affecting versions from 3.3.0 up to but not including 4.0.17. The flaw occurs during BMP header reconstruction when processing ICO images with a crafted small data buffer, causing heap corruption and a deterministic worker crash. This results in denial of service by terminating the worker handling the request.
A vulnerability in imgproxy's ICO image loader allows unauthenticated attackers to trigger unbounded memory allocation by crafting a specially formed ICO file. The flaw exists because the ICO loader reads a data size field from the file without validation and uses it directly to allocate memory, potentially causing a ~4GB allocation from a small 23-byte file. This can lead to denial of service via memory exhaustion on affected imgproxy versions prior to 4.0.16.
The anacrolix/torrent library's bencode decoder lacks any limit on the nesting depth of dictionaries and lists, leading to two denial-of-service vulnerabilities. First, deeply nested lists in a .torrent file can cause a stack overflow crash due to unbounded recursion, terminating the process. Second, the BitTorrent v2 file tree parser re-parses subtrees recursively in a quadratic time complexity manner, causing CPU exhaustion and process hang. These issues affect all applications embedding this library that process untrusted torrent metadata, including CLI tools and clients accepting magnet-link/v2 metadata. A fix is available in version 1.61.1-0.20260911233437-76452a2c8a2f.
CVE-2026-96634 is a remote code execution vulnerability in AdminNeo's SQLite functionality. It arises from a patch bypass exploiting catastrophic backtracking in a regular expression designed to filter ATTACH SQL statements. The vulnerability allows an attacker to bypass the ATTACH filter by causing PHP's preg_match() to fail due to excessive backtracking, enabling execution of malicious ATTACH queries that create PHP files on the server. This flaw affects AdminNeo versions prior to 5.5.1 and has a medium severity rating with a CVSS score of 6.9. A fix is available in version 5.5.1.
AdminNeo versions prior to 5.5.1 contain a vulnerability allowing authenticated SQLite users to delete arbitrary files writable by the PHP process via the database-list drop action. The flaw arises because the drop operation does not validate file extensions or restrict deletions to known database files, unlike create and rename operations. An attacker with valid credentials and a CSRF token can submit arbitrary file paths to be unlinked, potentially deleting critical files. This vulnerability is fixed in version 5.5.1.
AdminNeo versions 4.6.0 through 5.5.0 are vulnerable to cookie attribute injection via the unsanitized X-Forwarded-Prefix HTTP header. This allows an attacker who can control or influence this header to inject arbitrary cookie attributes into all Set-Cookie response headers, including session and encryption key cookies. The vulnerability enables session hijacking, credential theft, and session fixation by manipulating cookie scope and attributes. The issue arises from concatenating the X-Forwarded-Prefix header directly into the REQUEST_URI without sanitization and manually constructing Set-Cookie headers without escaping. A fix is available in AdminNeo version 5.5.1.
AdminNeo versions prior to 5.5.1 are vulnerable to a cross-site scripting (XSS) attack due to unsanitized insertion of the MySQL server version string into a script tag with a valid Content Security Policy (CSP) nonce. An attacker controlling a reachable MySQL server can craft a malicious version string that breaks out of the JavaScript context and executes arbitrary code in the victim's browser. This vulnerability requires no authentication when combined with login CSRF. A fix is available in version 5.5.1.
AdminNeo versions prior to 5.5.1 contain a vulnerability in their SQLite interface that allows authenticated users to execute arbitrary code. The application blocks ATTACH queries to prevent arbitrary file creation, but does not block the VACUUM INTO command, which can write the entire database to an arbitrary file path. An attacker can exploit this by injecting PHP code into a table name and then using VACUUM INTO to write a web shell to the web root, enabling remote command execution.