Exploit Intelligence Platform

CairoSVG is an SVG converter based on Cairo, a 2D graphics library. Prior to Kozea/CairoSVG has exponential denial of service via recursive <use> element amplification in cairosvg/defs.py. This causes CPU exhaustion from a small input.

eml_parser serves as a python module for parsing eml files and returning various information found in the e-mail as well as computed information. Prior to version 2.0.1, the official example script examples/recursively_extract_attachments.py contains a path traversal vulnerability that allows arbitrary file write outside the intended output directory. Attachment filenames extracted from parsed emails are directly used to construct output file paths without any sanitization, allowing an attacker-controlled filename to escape the target directory. This issue has been patched in version 2.0.1.

Stored Cross-Site Scripting (XSS) in the _genai/_evals_visualization component of Google Cloud Vertex AI SDK (google-cloud-aiplatform) versions from 1.98.0 up to (but not including) 1.131.0 allows an unauthenticated remote attacker to execute arbitrary JavaScript in a victim's Jupyter or Colab environment via injecting script escape sequences into model evaluation results or dataset JSON data.

MindsDB is a platform for building artificial intelligence from enterprise data. Prior to version 25.9.1.1, there is a path traversal vulnerability in Mindsdb's /api/files interface, which an authenticated attacker can exploit to achieve remote command execution. The vulnerability exists in the "Upload File" module, which corresponds to the API endpoint /api/files. Since the multipart file upload does not perform security checks on the uploaded file path, an attacker can perform path traversal by using `../` sequences in the filename field. The file write operation occurs before calling clear_filename and save_file, meaning there is no filtering of filenames or file types, allowing arbitrary content to be written to any path on the server. Version 25.9.1.1 patches the issue.

ChatterBot is a machine learning, conversational dialog engine for creating chat bots. ChatterBot versions up to 1.2.10 are vulnerable to a denial-of-service condition caused by improper database session and connection pool management. Concurrent invocations of the get_response() method can exhaust the underlying SQLAlchemy connection pool, resulting in persistent service unavailability and requiring a manual restart to recover. Version 1.2.11 fixes the issue.

In the query parser in OpenStack Vitrage before 12.0.1, 13.0.0, 14.0.0, and 15.0.0, a user allowed to access the Vitrage API may trigger code execution on the Vitrage service host as the user the Vitrage service runs under. This may result in unauthorized access to the host and further compromise of the Vitrage service. All deployments exposing the Vitrage API are affected. This occurs in _create_query_function in vitrage/graph/query.py.

MLflow Use of Default Password Authentication Bypass Vulnerability. This vulnerability allows remote attackers to bypass authentication on affected installations of MLflow. Authentication is not required to exploit this vulnerability. The specific flaw exists within the basic_auth.ini file. The file contains hard-coded default credentials. An attacker can leverage this vulnerability to bypass authentication and execute arbitrary code in the context of the administrator. Was ZDI-CAN-28256.

An issue in OpenFUN Richie (LMS) in src/richie/apps/courses/api.py. The application used the non-constant time == operator for HMAC signature verification in the sync_course_run_from_request function. This allows remote attackers to forge valid signatures and bypass authentication by measuring response time discrepancies

yt-dlp is a command-line audio/video downloader. Starting in version 2023.06.21 and prior to version 2026.02.21, when yt-dlp's `--netrc-cmd` command-line option (or `netrc_cmd` Python API parameter) is used, an attacker could achieve arbitrary command injection on the user's system with a maliciously crafted URL. yt-dlp maintainers assume the impact of this vulnerability to be high for anyone who uses `--netrc-cmd` in their command/configuration or `netrc_cmd` in their Python scripts. Even though the maliciously crafted URL itself will look very suspicious to many users, it would be trivial for a maliciously crafted webpage with an inconspicuous URL to covertly exploit this vulnerability via HTTP redirect. Users without `--netrc-cmd` in their arguments or `netrc_cmd` in their scripts are unaffected. No evidence has been found of this exploit being used in the wild. yt-dlp version 2026.02.21 fixes this issue by validating all netrc "machine" values and raising an error upon unexpected input. As a workaround, users who are unable to upgrade should avoid using the `--netrc-cmd` command-line option (or `netrc_cmd` Python API parameter), or they should at least not pass a placeholder (`{}`) in their `--netrc-cmd` argument.

Werkzeug is a comprehensive WSGI web application library. Versions 3.1.5 and below, the safe_join function allows Windows device names as filenames if preceded by other path segments. This was previously reported as GHSA-hgf8-39gv-g3f2, but the added filtering failed to account for the fact that safe_join accepts paths with multiple segments, such as example/NUL. The function send_from_directory uses safe_join to safely serve files at user-specified paths under a directory. If the application is running on Windows, and the requested path ends with a special device name, the file will be opened successfully, but reading will hang indefinitely. This issue has been fixed in version 3.1.6.

Deserialization of untrusted data in Azure SDK allows an unauthorized attacker to execute code over a network.

NiceGUI is a Python-based UI framework. Prior to 3.7.0, NiceGUI's FileUpload.name property exposes client-supplied filename metadata without sanitization, enabling path traversal when developers use the pattern UPLOAD_DIR / file.name. Malicious filenames containing ../ sequences allow attackers to write files outside intended directories, with potential for remote code execution through application file overwrites in vulnerable deployment patterns. This design creates a prevalent security footgun affecting applications following common community patterns. Note: Exploitation requires application code incorporating file.name into filesystem paths without sanitization. Applications using fixed paths, generated filenames, or explicit sanitization are not affected. This vulnerability is fixed in 3.7.0.

Cybersecurity AI (CAI) is a framework for AI Security. In versions up to and including 0.5.10, the CAI (Cybersecurity AI) framework contains multiple argument injection vulnerabilities in its function tools. User-controlled input is passed directly to shell commands via `subprocess.Popen()` with `shell=True`, allowing attackers to execute arbitrary commands on the host system. The `find_file()` tool executes without requiring user approval because find is considered a "safe" pre-approved command. This means an attacker can achieve Remote Code Execution (RCE) by injecting malicious arguments (like -exec) into the args parameter, completely bypassing any human-in-the-loop safety mechanisms. Commit e22a1220f764e2d7cf9da6d6144926f53ca01cde contains a fix.

Llama Stack (aka llama-stack) before 0.4.0rc3 does not censor the pgvector password in the initialization log.

wheel is a command line tool for manipulating Python wheel files, as defined in PEP 427. In versions 0.40.0 through 0.46.1, the unpack function is vulnerable to file permission modification through mishandling of file permissions after extraction. The logic blindly trusts the filename from the archive header for the chmod operation, even though the extraction process itself might have sanitized the path. Attackers can craft a malicious wheel file that, when unpacked, changes the permissions of critical system files (e.g., /etc/passwd, SSH keys, config files), allowing for Privilege Escalation or arbitrary code execution by modifying now-writable scripts. This issue has been fixed in version 0.46.2.

vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.14.0, vLLM loads Hugging Face `auto_map` dynamic modules during model resolution without gating on `trust_remote_code`, allowing attacker-controlled Python code in a model repo/path to execute at server startup. An attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve arbitrary code execution on the vLLM host during model load. This happens before any request handling and does not require API access. Version 0.14.0 fixes the issue.

Docling Core (or docling-core) is a library that defines core data types and transformations in the document processing application Docling. A PyYAML-related Remote Code Execution (RCE) vulnerability, namely CVE-2020-14343, is exposed in docling-core starting in version 2.21.0 and prior to version 2.48.4, specifically only if the application uses pyyaml prior to version 5.4 and invokes `docling_core.types.doc.DoclingDocument.load_from_yaml()` passing it untrusted YAML data. The vulnerability has been patched in docling-core version 2.48.4. The fix mitigates the issue by switching `PyYAML` deserialization from `yaml.FullLoader` to `yaml.SafeLoader`, ensuring that untrusted data cannot trigger code execution. Users who cannot immediately upgrade docling-core can alternatively ensure that the installed version of PyYAML is 5.4 or greater.

pypdf is a free and open-source pure-python PDF library. An attacker who uses an infinite loop vulnerability that is present in versions prior to 6.6.2 can craft a PDF which leads to an infinite loop. This requires accessing the outlines/bookmarks. This has been fixed in pypdf 6.6.2. If projects cannot upgrade yet, consider applying the changes from PR #3610 manually.

Langflow is a tool for building and deploying AI-powered agents and workflows. Prior to version 1.7.0.dev45, multiple critical API endpoints in Langflow are missing authentication controls. The issue allows any unauthenticated user to access sensitive user conversation data, transaction histories, and perform destructive operations including message deletion. This affects endpoints handling personal data and system operations that should require proper authorization. Version 1.7.0.dev45 contains a patch.

AstrBot is a large language model chatbot and development framework. A path traversal vulnerability present in versions 3.4.4 through 3.5.12 may lead to information disclosure, such as API keys for LLM providers, account passwords, and other sensitive data. The vulnerability has been addressed in Pull Request #1676 and is included in version 3.5.13. As a workaround, users can edit the `cmd_config.json` file to disable the dashboard feature as a temporary workaround. However, it is strongly recommended to upgrade to version v3.5.13 or later to fully resolve this issue.