🚨 CVE-2026-100635
SiYuan before v3.8.4 contains an authentication bypass vulnerability in the publish service where session cookies are issued without Secure or SameSite attributes over plaintext HTTP connections. An on-path attacker can observe a valid publish-visitor-session-id cookie from a Basic Auth exchange and replay it to access authenticated publish endpoints without knowing the account password.

🎖@cveNotify

🚨 CVE-2026-100636
SiYuan versions before v3.8.4 contain a path traversal vulnerability in the exportBrowserHTML endpoint that allows authenticated administrators to write arbitrary HTML content to index.html outside the workspace directory. Attackers can supply a folder parameter with directory traversal sequences to escape the export directory and overwrite index.html in any pre-existing kernel-writable location, enabling stored XSS or workspace defacement.

🎖@cveNotify

🚨 CVE-2026-100637
SiYuan versions before v3.8.4 contain a path traversal vulnerability in the checkoutRepo endpoint that allows authenticated administrators to write JSON files outside the workspace. Attackers can supply a sessionID parameter containing directory traversal sequences to overwrite arbitrary JSON files in pre-existing kernel-writable directories outside workspace boundaries.

🎖@cveNotify

🚨 CVE-2026-100638
SiYuan versions before v3.8.4 contain a path traversal vulnerability in the setNotebookIcon endpoint that allows authenticated administrators to create arbitrary directory trees and write files outside the workspace boundary. Attackers can supply directory traversal sequences in the notebook parameter to escape the workspace data directory and write conf.json files to arbitrary locations accessible by the kernel process.

🎖@cveNotify

🚨 CVE-2026-100639
SiYuan v3.8.3 fails to HTML-escape the data-subtype attribute when generating gutter-button markup (app/src/protyle/gutter/button.ts, assigned via innerHTML in app/src/protyle/gutter/index.ts) from content pasted as plain-text Markdown containing a Kramdown inline attribute list (IAL). Because the shared Lute renderer parses Kramdown IAL from text/plain input, an attacker-supplied Markdown snippet using entity-encoded quotes in data-subtype breaks out of the attribute value when the gutter markup is re-parsed by the browser, injecting additional attributes such as autofocus and onfocus. If a victim pastes the crafted Markdown and the affected gutter control receives focus, the injected handler executes; in the Electron desktop application, where the main BrowserWindow enables Node integration and disables context isolation, this results in JavaScript execution with renderer Node.js privileges (remote code execution). Fixed in v3.8.4.

🎖@cveNotify

🚨 CVE-2026-100640
SiYuan before v3.8.4 contains an authorization omission in the siyuan-get IPC handler that allows remote-kernel renderers to access native clipboard formats by invoking clipboardReadMathML, clipboardReadOffice, and clipboardReadWPS commands with matching plaintext. Attackers controlling remote renderer content can obtain MathML formulas, Office bytes, and WPS bytes from local clipboard during user-mediated paste operations.

🎖@cveNotify

🚨 CVE-2026-100641
SiYuan before v3.8.4 does not HTML-escape stored flashcard block content before interpolating it into the card-manager list markup. Block content returned by /api/riff/getRiffCards is inserted into a card item template in app/src/card/viewCards.ts and assigned to listElement.innerHTML, so content such as <img src=invalid onerror=...> becomes an executable event-handler attribute. Because the SiYuan desktop (Electron) main window is created with nodeIntegration enabled and contextIsolation disabled, an administrator who opens the card manager on a workspace containing attacker-supplied flashcard content (for example introduced through contribution or import) executes the attacker's script in a privileged renderer, which can lead to arbitrary code execution on the host. The affected endpoint remains behind authentication and administrator-role checks; this is an untrusted-content-to-privileged-renderer issue, not an authorization bypass.

🎖@cveNotify

🚨 CVE-2026-100642
SiYuan versions from v2.1.0 before v3.8.4 contain a cross-site request forgery vulnerability in the CheckAuth lock-screen pass-through branch that grants administrator access to loopback requests without validating Origin headers. Attackers can craft malicious web pages that force victims to terminate the kernel process, read workspace configuration and proxy settings, and trigger administrative actions via zero-credential cross-origin requests from the victim's browser.

🎖@cveNotify

🚨 CVE-2026-100643
SiYuan versions before v3.8.4 fail to properly escape four stored Attribute View values in textarea elements, allowing authenticated attackers to inject JavaScript by modifying field descriptions, template sources, select option descriptions, or footer calculation templates. Attackers can execute stored JavaScript when other users open affected database menus, and in the Electron desktop app with nodeIntegration enabled, this leads to command execution with SiYuan process privileges.

🎖@cveNotify

🚨 CVE-2026-100644
SiYuan before v3.8.4 contains a SQL injection vulnerability in the graph query endpoint where the dailyNoteSavePath parameter is concatenated into SQL without escaping. Unauthenticated attackers on published sites with auth disabled can inject SQL via UNION SELECT to extract arbitrary database rows from all notebooks.

🎖@cveNotify

🚨 CVE-2026-100645
SiYuan versions 3.7.0 before 3.8.4 contain a stored cross-site scripting vulnerability in gallery and kanban database renderers where field descriptions are not escaped in aria-label attributes. In the Electron desktop app with nodeIntegration enabled, attackers can inject JavaScript that calls Node.js child_process APIs to execute arbitrary commands with user privileges.

🎖@cveNotify

🚨 CVE-2026-100646
SiYuan is a self-hosted personal knowledge management system. In versions up to and including 3.8.3, the kernel's authentication guards (CheckAuth in kernel/model/session.go and IsSessionOriginAllowed in kernel/util/net.go) fail open when the HTTP Origin header is absent, on the incorrect assumption that any browser-initiated cross-site request carries an Origin. Because browsers omit Origin on cross-site top-level GET navigations and no-cors GET subresource loads — and the session cookie is SameSite=Lax — a single cross-site GET issued from any attacker-controlled web page is granted RoleAdministrator, both on default installations with no access-authorization code and on password-protected instances with a live session. Combined with content-type sniffing on the /api/network/proxy endpoint, which allows attacker-controlled HTML to be served under SiYuan's own origin, this permits an unauthenticated remote attacker to execute arbitrary script in the SiYuan origin (http://127.0.0.1:6806), invoke administrator APIs, and exfiltrate the persistent kernel API token. This issue is fixed in version 3.8.4.

🎖@cveNotify

🚨 CVE-2026-100647
vLLM versions before 0.29.0 contain a denial-of-service vulnerability in the cache_salt parameter accepted on OpenAI-compatible and Anthropic API endpoints, which lacks maximum length validation and is processed on the single EngineCore scheduler thread. Unauthenticated attackers can send HTTP requests with multi-hundred-megabyte salt values that trigger expensive pickle serialization and SHA-256 hashing, stalling the scheduler thread and denying service to all concurrent requests.

🎖@cveNotify

🚨 CVE-2026-100648
vllm before 0.29.0 fails to enforce VLLM_MAX_AUDIO_CLIP_FILESIZE_MB limit in multimodal chat audio decoding, allowing unauthenticated clients to bypass file size restrictions. Attackers can submit oversized audio files through chat endpoints to consume excessive memory and CPU resources during decoding.

🎖@cveNotify

🚨 CVE-2026-100649
vLLM before 0.29.0 contains a resource-limit bypass vulnerability in PyNvVideoCodec decoder allocation where sampler subclass shadowing allows independent counter increments. Unauthenticated attackers can select different sampler subclasses in video requests to exceed configured decoder limits and exhaust unaccounted GPU memory.

🎖@cveNotify

🚨 CVE-2026-100650
vLLM through 0.29.0 fetches and fully materializes remote or inline media before enforcing its documented media controls (the VLLM_MAX_AUDIO_CLIP_FILESIZE_MB compressed-audio size cap, default 25 MB, and the per-modality --limit-mm-per-prompt item limits). Across four ingress paths — the shared media-acquisition layer (HTTPConnection.get_bytes()/async_get_bytes()), the chat completions audio_url/base64 path, the batch speech runner, and the Rust frontend POST /tokenize route — the server reads the entire HTTP response body, base64-decodes the inline payload, or spawns one fetch/decode task per media part, and only then applies the limit (or, on some paths, never applies it). A remote attacker can therefore cause the API server or batch-runner process to allocate memory and consume outbound bandwidth proportional to an attacker-chosen body size or media item count before the request is rejected, resulting in pre-inference memory and bandwidth exhaustion (denial of service). The chat and batch surfaces require an API key when one is configured; the Rust frontend /tokenize route is unauthenticated by design. There is no code execution or data disclosure impact.

🎖@cveNotify

🚨 CVE-2026-100651
vLLM before 0.29.0 fails to enforce decoder prompt-length validation on the disaggregated serving endpoint /inference/v1/generate. When the request contains a 'features' (multimodal) payload, vllm/entrypoints/serve/disagg/serving.py builds a multimodal EngineInput directly from the caller-supplied token_ids, and GenerateRequest.token_ids (vllm/entrypoints/serve/disagg/protocol.py) is not checked against model_config.max_model_len. For multimodal processors that report skip_prompt_length_check=True (for example Nemotron Parse, Whisper, and FireRedLID), InputProcessor._validate_prompt_len() returns immediately for both encoder and decoder prompts, so an overlong prompt becomes an EngineCoreRequest and reaches the worker input-batch copy into a fixed max_model_len-wide NumPy row. A client able to reach the endpoint on an affected model configuration can therefore submit an overlong token_ids list to trigger a worker failure and denial of service. Fixed in 0.29.0.

🎖@cveNotify

🚨 CVE-2026-100652
vLLM versions 0.22.0 through 0.23.0 fail to validate stop_token_ids against vocabulary bounds in Rust HTTP and gRPC frontends, allowing out-of-vocabulary token IDs to reach MinTokensLogitsProcessor. Attackers can submit requests with min_tokens greater than zero and out-of-vocabulary stop_token_ids to trigger CUDA tensor indexing failures that leave EngineCore in a fatal state requiring service restart.

🎖@cveNotify

🚨 CVE-2026-100653
vLLM is an inference and serving engine for large language models. In versions from 0.22.1 through 0.28.0, the operator-supplied model revision pin (--revision / --code-revision) is not propagated to several Hugging Face artifact loads for the FunAudioChat and Tarsier2 architectures: the WhisperFeatureExtractor and speech_tokenizer PreTrainedTokenizerFast loads in vllm/model_executor/models/funaudiochat.py and the Qwen2VLConfig.from_pretrained call used by Tarsier2ProcessingInfo in vllm/model_executor/models/qwen2_vl.py. As a result, deployments pinned to a reviewed revision still resolve these behavior-affecting processor, tokenizer, and config artifacts from the repository's default revision, so a later change to the upstream default branch can alter audio preprocessing, speech tokenizer behavior, or Tarsier2 configuration without any change to the operator's configured pin. This is a supply-chain integrity and reproducibility failure for pinned deployments; it is residual to the earlier fix tracked as GHSA-3ww4-5jv9-j5gm / CVE-2026-47155 and does not constitute remote code execution or a trust_remote_code=False bypass. The issue is fixed in version 0.28.0.

🎖@cveNotify

🚨 CVE-2026-100654
vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens > 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).

🎖@cveNotify

🚨 CVE-2026-100656
Netty (io.netty:netty-codec-http) contains an unbounded per-connection queue growth flaw in HttpServerCodec. The codec tracks the HTTP method of each still-unanswered pipelined request; the first 32 entries are bit-packed into a single long, but every additional entry is appended to methodOverflowQueue, an ArrayDeque with no size limit and no rejection path. A remote, unauthenticated attacker who pipelines HTTP/1.1 requests on a single connection while withholding reads on their own end (preventing responses from being flushed) can grow this queue without bound, causing unbounded heap growth and denial of service. Affected versions are 4.2.0.Final through 4.2.17.Final and all releases up to and including 4.1.137.Final; the issue is fixed in 4.2.18.Final and 4.1.138.Final.

🎖@cveNotify