Vulnerabilities (CVE)

AIOHTTP is an asynchronous HTTP client/server framework for asyncio and Python. Prior to 3.14.1, DigestAuthMiddleware can send an authentication response after following a cross-origin redirect. This likely requires an open redirect vulnerability or similar on the target domain for an attacker to be able to execute. Further, the attacker is only receiving the digest, so should only be able to extract the user's credentials if the cryptography is weak or there is some kind of password reuse. This vulnerability is fixed in 3.14.1.Show less

AIOHTTP is an asynchronous HTTP client/server framework for asyncio and Python. Prior to 3.14.1, the server_hostname TLS SNI check can be bypassed when an existing connection is reused. If an application makes multiple requests to the same domain, but with different per-request server_hostname parameters, then the later calls may succeed by reusing the existing connection when they should have been rejected due to the TLS SNI check. This vulnerability is fixed in 3.14.1.Show less

AIOHTTP is an asynchronous HTTP client/server framework for asyncio and Python. Prior to 3.14.1, if an attacker sends large incomplete websocket frame payloads, it may be possible to bypass the usual size limits on memory use. This vulnerability is fixed in 3.14.1.Show less

AIOHTTP is an asynchronous HTTP client/server framework for asyncio and Python. Prior to 3.14.1, no limit was present on the number of pipelined requests that could be queued. An attacker may be able to use pipelined requests to use excessive amounts of memory, potentially leading to DoS. This vulnerability is fixed in 3.14.1.Show less

protobufjs-cli is the command line add-on for protobuf.js. Prior to 1.3.2 and 2.5.0, a previous fix for unsafe name handling in pbjs static / static-module code generation was incomplete. Affected versions of protobufjs-cli could still emit unsafe JavaScript references when generating static output from crafted JSON descriptor input. The common case of parsing schemas from .proto files is not affected. This is a bypass of CVE-2026-44295. An attacker who can provide or influence pre-parsed JSON descriptors passed to pbjs static code generation may be able to cause generated JavaScript output to contain attacker-controlled code. The injected code may execute if the generated file is later executed or imported and an affected generated API path is invoked. This vulnerability is fixed in 1.3.2 and 2.5.0.Show less

protobufjs compiles protobuf definitions into JavaScript (JS) functions. From 8.2.0 to 8.4.2, protobufjs preserved unknown wire elements in message.$unknowns and did not provide a decode-time option to discard unknown fields before retaining them. A crafted protobuf payload containing many unknown fields could therefore cause a decoded message to retain substantially more memory than the input size would suggest, even when unknown-field round-tripping is not needed. protobufjs 8.5.0 added the relevant decode-time options, allowing applications that decode untrusted protobuf data to disable unknown-field retention during decode. protobufjs 8.6.2 flips the default so unknown fields are discarded unless explicitly opted into.Show less

protobufjs compiles protobuf definitions into JavaScript (JS) functions. Prior to 8.6.0 and 7.6.3, protobufjs accepted certain schema-derived names that could collide with properties used by protobufjs runtime helpers. The known affected names are fields named hasOwnProperty, field or oneof names such as $type when loaded through protobufjs JSON/reflection descriptors, and service methods whose generated helper name is rpcCall. When affected message or service types were used, protobufjs could read schema-controlled data where it expected an own-property helper, reflected type metadata, or the base RPC helper. This could cause deterministic exceptions or recursive calls in affected decode post-checks, verification, object conversion, reflected JSON serialization, or protobufjs RPC helper invocation. This vulnerability is fixed in 8.6.0 and 7.6.3.Show less

vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.23.1rc0, the fix for CVE-2026-22778, which introduced a sanitize_message helper that strips object-repr memory addresses from error messages before they reach the client, is incomplete: several response paths echo str(exc) directly to clients without calling sanitize_message. The unsanitized sites include the Anthropic API router in vllm/entrypoints/anthropic/api_router.py (the POST /v1/messages and POST /v1/messages/count_tokens handlers), the Server-Sent Events streaming converter in vllm/entrypoints/anthropic/serving.py, and the realtime speech-to-text WebSocket in vllm/entrypoints/speech_to_text/realtime/connection.py. These paths catch the exception inside the route coroutine and construct the JSONResponse themselves, bypassing the sanitizing global FastAPI exception handler, and WebSocket frames do not traverse that handler chain at all. Using the same primitive as the parent issue, an unauthenticated attacker can send malformed image bytes through the Anthropic Messages API image content parts so that PIL.Image.open raises an UnidentifiedImageError whose message contains the BytesIO object repr, leaking the heap memory address verbatim in the error.message field of the response body. This vulnerability is fixed in 0.23.1rc0.Show less

vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.23.1rc0, ll temperature validation gates use comparison operators (<, >), which silently evaluate to False for NaN and for positive Infinity in Python's IEEE 754 float semantics. Both values pass every guard and propagate to GPU sampling kernels, where they produce undefined behavior or CUDA errors that can crash the inference worker. This vulnerability is fixed in 0.23.1rc0.Show less

vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.23.1rc0, vLLM's /v1/audio/transcriptions endpoint limits compressed upload size but not decoded PCM output. A 25MB OPUS file expands to ~14.9GB of float32 PCM at decode time. This vulnerability is fixed in 0.23.1rc0.Show less

vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.22.1, the vLLM Dockerfile is vulnerable to a dependency confusion attack through the flashinfer-jit-cache package. The package is installed from a custom index (flashinfer.ai/whl/) using --extra-index-url, but the package name was not registered on PyPI, and UV_INDEX_STRATEGY="unsafe-best-match" is set globally. An attacker who registers flashinfer-jit-cache on PyPI with version 0.6.11.post2 can execute arbitrary code as root during the Docker build and backdoor every resulting container image, enabling exfiltration of all user prompts, API credentials, and model data from production vLLM deployments This vulnerability is fixed in 0.22.1.Show less

UBB.threads is vulnerable to Stored XSS via user posts and user profile fields. The application fails to properly sanitize user input, allowing low privileged attackers to inject arbitrary JavaScript that executes in a victim's browser upon viewing. Because vendor contact attempts were unsuccessful, the vulnerability has only been confirmed in version 7.7.5 but may also affect other versions.Show less

SiYuan is an open-source personal knowledge management system. Prior to 3.7.0, the attribute-view (database) cell renderer genAVValueHTML interpolates cell content raw in four of its branches: text, url, phone, and mAsset. A cell value like </textarea><img src=x onerror="..."> or "><img src=x onerror="..."> breaks out of its surrounding tag and runs arbitrary JavaScript in the renderer when the victim opens the block-attribute panel. On Electron desktop the renderer runs with nodeIntegration:true, so the XSS chains to host RCE via require('child_process'). AV files live under the workspace and ride normal sync, so an attacker with write access to any synced workspace plants the payload once and it fires on every device that opens a panel containing that row.he kernel doesn't escape on the way in either, so the malicious cell persists byte-for-byte. There's no equivalent of the html.EscapeAttrVal call that protects block IAL attributes at kernel/model/blockial.go:261. This vulnerability is fixed in 3.7.0.Show less

jmespath.php allows users to use JMESPath, software for declaratively specifying how to extract elements from a JSON document, in PHP applications with PHP data structures. Versions prior to 2.9.1 can generate and execute attacker-controlled PHP code when `JmesPath\CompilerRuntime` is used with an attacker-controlled JMESPath expression. The compiler emits parsed JMESPath function names into generated PHP source without sufficient escaping. A crafted expression can cause the generated cache file to contain executable attacker-controlled PHP, which is then loaded by the compiler runtime. The issue is patched in `2.9.1` and later. As a workaround, disable `JP_PHP_COMPILE` and do not use `JmesPath\CompilerRuntime` with attacker-controlled expressions. Use the default `AstRuntime` for untrusted expressions. Applications that must continue accepting untrusted JMESPath expressions before upgrading should ensure those expressions are never evaluated by the compiler runtime.Show less

SiYuan is an open-source personal knowledge management system. Prior to 3.7.0, renderPackageREADME in kernel/bazaar/readme.go renders a Bazaar package README from Markdown to HTML with the lute engine and SetSanitize(true). The lute sanitizer is an event-handler blocklist: allowAttr rejects only attribute names present in a fixed eventAttrs map copied from the w3schools legacy handler list. That map omits modern event handlers. onpointerover, onpointerdown, onauxclick, onbeforetoggle, onfocusin, onanimationstart, and ontransitionend are not in the list, so the sanitizer passes them through verbatim on any tag. The frontend assigns the rendered HTML to mdElement.innerHTML in app/src/config/bazaar.ts with no client-side DOMPurify on this path, into a normal element in the main document (no iframe, no sandbox). The kernel sends no Content-Security-Policy, X-Frame-Options, or X-Content-Type-Options header on any response, so an inline handler runs when its event fires. The README is rendered when an Administrator opens a package in Settings → Marketplace, after the one-time marketplace trust consent. Install is not required. Result: a third-party Bazaar package author runs JavaScript in the Administrator's authenticated SiYuan origin when the Administrator views and interacts with the package listing, and gains full control of the workspace. This vulnerability is fixed in 3.7.0.Show less

SiYuan is an open-source personal knowledge management system. Prior to 3.7.0, SiYuan Note's kernel HTTP server unconditionally trusts all chrome-extension:// origins, granting RoleAdministrator access to every installed browser extension without any authentication. Combined with the default empty AccessAuthCode on desktop installs, any Chrome/Chromium extension -- including a compromised legitimate extension via supply chain attack -- can make fully authenticated admin API calls to the SiYuan kernel at 127.0.0.1:6806, enabling data exfiltration, stored XSS injection, and configuration tampering. This vulnerability is fixed in 3.7.0.Show less

SiYuan is an open-source personal knowledge management system. Prior to 3.7.0, the /api/icon/getDynamicIcon endpoint is explicitly excluded from authentication in SiYuan's kernel router (router.go, "不需要鉴权" -- no auth needed). When called with type=8 and a valid block id parameter, this endpoint invokes RenderDynamicIconContentTemplate, which executes a Go template that includes the querySQL and queryBlocks functions. These functions run arbitrary SELECT statements against the SiYuan SQLite database. An unauthenticated network-adjacent attacker who knows a valid block ID can exfiltrate all user note content, tags, asset references, and block attributes from the database. This vulnerability is fixed in 3.7.0.Show less

SiYuan is an open-source personal knowledge management system. Prior to 3.7.0, CSS snippet body containing </style> breaks out of its surrounding <style> tag when renderSnippet() interpolates it via insertAdjacentHTML. A payload like runs arbitrary JavaScript in the renderer. On Electron desktop builds the renderer runs with nodeIntegration:true, so require('child_process') is reachable from the injected handler and the XSS chains to host RCE. Snippets sync via the workspace repository, so an attacker with write access to any synced workspace plants the payload once and it fires on every device that pulls. The bug also bypasses the user's enabledCSS / enabledJS separation. A user who turned enabledJS off was making a deliberate call not to run untrusted JavaScript; the CSS path runs it anyway. This vulnerability is fixed in 3.7.0.Show less

SiYuan is an open-source personal knowledge management system. Prior to 3.7.0, the patch for CVE-2026-41894 ("Path Traversal via Double URL Encoding") sanitized the /export/ route but the identical root cause remains in the /assets/*path route. In publish mode (anonymous read-only HTTP endpoint, default port 6808), an unauthenticated remote attacker can read arbitrary files inside WorkspaceDir — including conf/conf.json (which contains the AccessAuthCode SHA256 hash, API token, and sync keys), temp/siyuan.db, temp/blocktree.db, and siyuan.log — by double-URL-encoding .. segments. This vulnerability is fixed in 3.7.0.Show less

Kitty is a cross-platform GPU based terminal. In versions prior to 0.47.3, kitty's OSC 21 (color-control) query reply reflects attacker-controlled bytes, including newlines, into the shell's input without sanitization. Version 0.47.3 fixes the issue.Show less

Kitty is a cross-platform GPU based terminal. In versions 0.47.0 and 0.47.1, `kitten dnd` can allow a malicious remote drag-and-drop source to overwrite or truncate arbitrary files writable by the local kitty user. Remote `text/uri-list` drops are staged in a temporary directory, but on case-sensitive filesystems duplicate remote basenames are not de-duplicated. An attacker can first create a staged symlink and then send a same-name regular-file entry. The regular-file write uses `utils.CreateAt()` / `openat(O_RDWR|O_CREAT|O_TRUNC)` without `O_NOFOLLOW`, so it follows the attacker-created symlink and writes outside the staging directory before final overwrite confirmation runs. This appears related in class to the file-transfer symlink advisory, but it is a different bug: it affects `kitten dnd` remote drag-and-drop staging, uses different vulnerable code (`kittens/dnd/drop.go` and `tools/utils/file_at_fd.go`), and reproduces on commit `4aa4a5c0567a92553a8c20a88a4352da637fca5d`, after the file-transfer `O_NOFOLLOW` fix. Version 0.47.2 patches the issue.Show less

It has been identified that a vulnerability (CWE-427) exists in the UPS (Uninterruptible Power Supply) management application, whereby improper permissions on the installation directory allow a malicious actor to place a DLL that is then executed with administrator privileges. If a malicious DLL is placed in the installation directory of this product, there is a possibility that the malicious DLL may be executed by exploiting the product’s behavior of loading missing DLLs from the same directory as the executable during service startup.Show less

vLLM is an inference and serving engine for large language models (LLMs). From 0.5.5 until 0.23.1rc0, integer truncation of tensor dimensions in vLLM's GGUF dequantize kernels (csrc/quantization/gguf/gguf_kernel.cu) causes partial tensor processing. The output tensor is allocated at full size via torch::empty (uninitialized memory), but the dequantize CUDA kernel processes only a truncated number of elements. The unfilled portion of the output tensor retains whatever was previously in GPU memory. In multi-tenant inference deployments, this residual GPU memory may contain tensor data from other users' inference requests, constituting information disclosure. This vulnerability is fixed in 0.23.1rc0.Show less