聚合 NVD、CVE 及多源情資，深度解析 RCE 等高危風險。系統整合 CVSS 與 EPSS 模型，動態追蹤 Exploit 資源與 PoC 公開狀態，研判可利用性。結合官方修補與修復方案，優化漏洞管理優先級，縮短回應週期，保障資產安全。
指派機構（CNA / 來源）：[email protected] 移除此篩選
| CVE | 描述 | 最高 CVSS | EPSS % | 公開時間 | 更新時間 |
|---|---|---|---|---|---|
| CVE-2026-12252 | In nltk/nltk versions 3.9.3 and earlier, five Stanford interface classes (StanfordPOSTagger, StanfordNERTagger, StanfordParser, StanfordDependencyParser, and StanfordNeuralDependencyParser) are vulnerable to untrusted JAR code execution. These classes accept user-controllable JAR paths and execute them via the `java()` function, which invokes `subprocess.Popen()` without integrity verification. This vulnerability is identical to CVE-2026-0848, which was fixed for StanfordSegmenter by adding SHA2 | 7.8 | 無 | 2026-07-03 | 2026-07-03 |
| CVE-2026-12481 | A vulnerability in keras-team/keras version 3.14.0 allows for arbitrary code execution due to improper handling of deserialization in the `Lambda` layer. Specifically, the `_raise_for_lambda_deserialization()` function fails to enforce the safe-mode guard when `safe_mode` is set to `None`, which is the default value when `from_config()` is called outside of a `SafeModeScope` context. This logic error conflates `None` (unset/default-deny) with `False` (explicitly disabled), bypassing the guard an | 8.8 | 無 | 2026-07-03 | 2026-07-03 |
| CVE-2026-8147 | In MLflow versions prior to 3.14.0, when running with authentication enabled, the trace API endpoints lack proper authorization validators. This allows any authenticated user to bypass experiment-level authorization controls on all trace operations, including reading, deleting, and modifying traces on experiments they do not have permission to access. The issue arises from the `_before_request` handler, which does not register authorization validators for trace endpoints, resulting in requests p | 8.1 | 0.34% | 2026-07-02 | 2026-07-02 |
| CVE-2026-12480 | Keras versions up to and including 3.13.2 are vulnerable to an arbitrary HDF5 file read due to an incomplete fix for CVE-2026-1669. The vulnerability resides in the `H5IOStore._verify_dataset()` and `file_editor.py` methods, which fail to check the `dataset.is_virtual` property of HDF5 datasets. This allows an attacker to craft a malicious `.keras` model archive or `.h5` weights file containing a Virtual Dataset (VDS) that references external HDF5 files on the victim's filesystem. When the victi | 5.5 | 0.13% | 2026-07-01 | 2026-07-02 |
| CVE-2026-8387 | A vulnerability in allegroai/clearml versions up to and including 1.16.5 allows for relative path traversal when extracting `.zip` archives using the `ZipFile.extractall()` method in `StorageManager._extract_to_cache()`. This issue arises due to the lack of path traversal validation, enabling an attacker to write arbitrary files to the filesystem. Attack vectors include dataset downloads, artifact downloads, model downloads, and offline session imports. The vulnerability can lead to remote code | 2.4 | 0.36% | 2026-07-01 | 2026-07-02 |
| CVE-2026-12243 | NLTK version 3.9.4 is vulnerable to a path traversal attack due to an incomplete fix for GitHub Issue #3504. The `_UNSAFE_NO_PROTOCOL_RE` regex in `nltk/data.py` checks for literal `../` sequences but fails to account for percent-encoded traversal sequences such as `..%2f`. The `url2pathname()` function decodes these sequences after the validation step, allowing an attacker to bypass the protection. This vulnerability enables an attacker to read arbitrary files accessible to the Python process b | 7.5 | 0.49% | 2026-06-29 | 2026-06-30 |
| CVE-2026-6658 | A vulnerability in jupyter/nbconvert versions <= 7.17.0 allows for Cross-site Scripting (XSS) via unsanitized `text/vnd.mermaid` output in HTML exports. The `data_mermaid` block in `share/templates/lab/base.html.j2` renders `text/vnd.mermaid` cell output directly into HTML without escaping, enabling attackers to inject arbitrary HTML/JavaScript by breaking out of the `<pre>` tag. This vulnerability impacts any server using nbconvert to render notebooks as HTML, allowing attackers to execute arbi | 5.4 | 0.13% | 2026-06-26 | 2026-06-26 |
| CVE-2026-12479 | A path traversal vulnerability exists in keras-team/keras version 3.14.0, specifically in the `DiskIOStore.make` method within the Keras 3 model saving and loading library. This vulnerability arises from the improper handling of user-provided layer names, which are used to construct directory paths without sanitizing for parent directory components (`..`). While forward slashes (`/`) are restricted in layer names, directory traversal sequences are not. This allows an attacker to craft a maliciou | 6.1 | 0.26% | 2026-06-22 | 2026-06-22 |
| CVE-2026-5366 | Prefect version 3.6.23 is vulnerable to remote code execution due to improper handling of user-controlled input in the `GitRepository` storage class. The `commit_sha` parameter, which is passed to git commands, lacks validation and does not include a `--` separator to distinguish user input from git flags. This allows attackers to inject arbitrary git flags, such as `--upload-pack`, enabling execution of external programs. Additionally, the `directories` parameter can be exploited to inject git | 9.9 | 0.87% | 2026-06-20 | 2026-07-01 |
| CVE-2026-9375 | urllib3 version 2.6.3 is vulnerable to a decompression bomb bypass in its streaming API (`preload_content=False`) when using Brotli support. The issue arises due to three independent code paths in `response.py` that bypass the `max_length` protection introduced in version 2.6.0 to mitigate CVE-2025-66471. Specifically, negative `max_length` values can be produced due to buffer arithmetic in `read()`, `flush_decoder` unconditionally overrides `max_length` to `-1`, and `_flush_decoder()` passes no | 7.5 | 0.30% | 2026-06-19 | 2026-06-22 |
| CVE-2026-12199 | A vulnerability in `nltk.app.wordnet_app` up to version 3.9.3 allows unauthenticated remote shutdown of the local WordNet Browser HTTP server when started in its default mode. The server listens on all interfaces and processes a specific unauthenticated GET request (`/SHUTDOWN%20THE%20SERVER`) to terminate the process immediately via `os._exit(0)`. This results in a denial of service, impacting service availability. The issue arises due to insufficient authentication and protection mechanisms fo | 7.5 | 0.33% | 2026-06-17 | 2026-06-17 |
| CVE-2026-3840 | A vulnerability in Kedro version 1.2.0 allows an attacker to exploit path traversal by providing a crafted version string. The `_get_versioned_path()` method in `kedro/io/core.py` directly interpolates user-supplied version strings into filesystem paths without sanitization. This enables an attacker to escape the intended versioned dataset directory and access files outside the expected path. The issue is also reachable through the CLI via the `--load-versions` parameter, as `_split_load_version | 7.1 | 0.19% | 2026-06-12 | 2026-06-17 |
| CVE-2026-11816 | Keras versions prior to 3.14.0 are vulnerable to a path traversal issue in the archive extraction utilities located in `keras/src/utils/file_utils.py`. The functions `filter_safe_tarinfos()` and `filter_safe_zipinfos()` validate archive member paths against the process current working directory (CWD) instead of the actual extraction destination. When the process runs with CWD set to `/`, which is common in Docker containers, CI/CD runners, and Jupyter environments, the validation boundary become | 8.1 | 0.52% | 2026-06-11 | 2026-07-01 |
| CVE-2026-5497 | vLLM versions 0.8.0 and later are vulnerable to an Out-of-Memory (OOM) Denial of Service (DoS) attack due to unbounded frame count processing in the `VideoMediaIO.load_base64()` method. When processing `video/jpeg` data URLs, the method splits the base64 data string on commas to extract individual JPEG frames without enforcing a frame count limit. An attacker can exploit this by crafting a single API request containing thousands of comma-separated base64-encoded JPEG frames in a data URL, causin | 7.5 | 0.60% | 2026-06-11 | 2026-07-03 |
| CVE-2026-6657 | A vulnerability in jupyter-server versions 1.12.0 through 2.17.0 allows an attacker to bypass CORS origin validation when the `allow_origin_pat` configuration is used. The issue arises from the use of `re.match()` for validating the `Origin` header, which only anchors at the start of the string. This allows attacker-controlled domains such as `trusted.example.com.evil.com` to pass validation against patterns intended to match `trusted.example.com`. The vulnerability affects multiple locations in | 8.8 | 0.20% | 2026-06-03 | 2026-06-30 |
| CVE-2026-5241 | A vulnerability in the LightGlue model loading path of huggingface/transformers version 5.2.0 allows an attacker-controlled model repository to execute arbitrary code during model initialization. The issue arises because the `trust_remote_code` parameter, intended to prevent remote code execution, is overridden by untrusted serialized configuration data in a nested code path. Specifically, when loading a LightGlue model using `AutoModel.from_pretrained()` with `trust_remote_code=False`, the `Lig | 9.6 | 0.49% | 2026-06-03 | 2026-07-02 |
| CVE-2026-4035 | A vulnerability in mlflow/mlflow versions prior to 3.11.0 allows for the resolution of environment variables in AI Gateway secrets, which can be exploited to exfiltrate sensitive server-side environment credentials to an attacker-controlled endpoint. This issue arises because the `api_key` field in gateway secrets can accept `$ENV_VAR` references, which are resolved against the MLflow server's environment during runtime. The resolved secrets are then sent in provider authentication headers to th | 7.7 | 0.38% | 2026-06-03 | 2026-06-29 |
| CVE-2026-5422 | A path traversal vulnerability exists in jupyter-server version 2.17.0 due to an incorrect root directory boundary check in the _get_os_path() function within jupyter_server/services/contents/fileio.py. The check uses startswith(root) without appending a trailing path separator, allowing sibling directories with names starting with the same prefix as root_dir to bypass the check. Additionally, the to_os_path() function in utils.py does not strip ".." from path parts, enabling traversal sequences | 8.1 | 0.44% | 2026-06-02 | 2026-06-17 |
| CVE-2026-3514 | In version 3.6.19 of prefecthq/prefect, an authentication bypass vulnerability exists due to the improper handling of URL path exemptions for health check probes. Specifically, the authentication middleware exempts any URL path ending with 'health' or 'ready' from authentication checks. This allows an attacker to create resources with names ending in 'health' or 'ready' and access them without authentication. Affected endpoints include those for variables, flows, work pools, work queues, and dep | 7.5 | 0.48% | 2026-06-02 | 2026-06-17 |
| CVE-2026-3198 | MLflow 3.9.0 with basic-auth (`--app-name basic-auth`) fails to enforce authorization checks for multiple Gateway API 'list' endpoints. Specifically, the `BEFORE_REQUEST_HANDLERS` dictionary in `mlflow/server/auth/__init__.py` does not include entries for `ListGatewaySecretInfos`, `ListGatewayEndpoints`, and `ListGatewayModelDefinitions`. This allows any authenticated user, regardless of their assigned permissions, to enumerate all gateway secrets, endpoints, and model definitions. This vulnerab | 6.5 | 0.21% | 2026-06-02 | 2026-06-17 |