T08 · Insecure Dependencies
Warning
- Location
README.md:9- Finding
Unpinned Third-Party Packages and Unverified AI Model Downloads
- Content
View full analysis
Vulnerability Details
File Location:
README.md:9-30; additional runtime model-loading guidance inSKILL.md:32-47
Vulnerability Type: T08: Insecure Dependencies
Risk Level: MediumVulnerable Code
bash **Windows:** 1. Download Ollama from https://ollama.com/download 2. Install and verify: `ollama --version` **Pull the model:** ```bash ollama pull qwen3-vl:8bmacOS/Linux:
bash brew install ollama # or download from ollama.com ollama pull qwen3-vl:8bAudio Recognition - FunASR
Create conda environment:
bash conda create -n asr-local python=3.11 conda activate asr-local pip install funasrtext The related model-loading instructions in `SKILL.md:32-47` are: ```bash conda run -n asr-local python -c " import os os.environ['MODELSCOPE_CACHE'] = 'C:/Users/TOM/.cache/modelscope' from funasr import AutoModel model = AutoModel( model='iic/speech_seaco_paraformer_large_asr_nat-zh-cn-16k-common-vocab8404-pytorch', model_revision='v2.0.4', disable_update=True, ncpu=4 ) result = model.generate(input='AUDIO_PATH') print(result) "bash ollama run qwen3-vl:8b "Describe this image in detail: /path/to/frame.jpg"The troubleshooting guidance in
README.md:58-61also permits changing the model source:bash **Model download slow:** ```bash # Use a mirror if needed OLLAMA_BASE_URL=https://example.com ollama pull qwen3-vl:8btext ### Technical Analysis The installation instructions use `pip install funasr` without an exact version, dependency lock file, or integrity hashes. Consequently, the installed FunASR version and its transitive dependency graph can change between installations. Python packages can execute package-controlled build or installation logic, so compromise of an upstream package, registry account, or dependency resolution process could result in code execution during installation or import. The Ollama and ModelScope artifacts are also identified by mutable names or rev ...[truncated 2173 chars]- Remediation
View remediation
Remediation Suggestions
- Pin FunASR and all transitive dependencies to reviewed versions using a lock file.
- Use hash-verified installation, such as a requirements file containing exact versions and
--hashentries, and install withpip install --require-hashes. - Record the supported Python, Conda, Ollama, ffmpeg, and model-runtime versions to make installations reproducible.
- Pin AI models to immutable manifests or cryptographic digests where the relevant tooling supports them.
- Publish trusted SHA-256 checksums or signed provenance for downloaded model artifacts and document verification before first use.
- Restrict downloads to explicitly approved official registries. Remove the generic mirror recommendation or replace it with a documented, authenticated allowlist of trusted mirrors.
- Perform package and model retrieval in a least-privileged environment without administrator rights.
- Isolate video processing in a dedicated container, virtual machine, or restricted operating-system account with minimal file and network access.
- Scan locked dependencies for known vulnerabilities and review updates before changing pinned versions.
- Document that the initial model download requires network access and provide an offline installation procedure for security-sensitive deployments.