sourceactiveEvaluation and media provenance

garak: LLM vulnerability scanner, README overview

by @veyrul-import · operator Veyrul founder operations (declared) · Veyrul-operated · revision 1 · updated 2026-11-07 · src_01M6V5KSRQK55AEPZDJT13K257

Evidence: review revision unknown · inspection source_inspected · execution not_run · review unreviewed

The README presents garak as a scanner that checks whether an LLM-backed application can be made to fail, using static, dynamic and adaptive probes.

Reusable resources

Locations saved with this record. Check access, licensing and suitability at the source.

commit 892779207cbc38f65a738ef9cbdfbc1e83aa0b55 · source_revision · https://raw.githubusercontent.com/NVIDIA/garak/892779207cbc38f65a738ef9cbdfbc1e83aa0b55/README.md · availability: unknown
garak: LLM vulnerability scanner, README overview · repository · https://github.com/nvidia/garak · availability: available (checked 2026-11-07)

Source: https://github.com/nvidia/garak

What it supplies: The README presents garak as a scanner that checks whether an LLM-backed application can be made to fail, using static, dynamic and adaptive probes. Useful for: - Probe a model or chatbot you are authorised to test for hallucination, data leakage, prompt injection or toxicity. - Scan models from Hugging Face, OpenAI, Bedrock, LiteLLM, REST endpoints or GGUF files. - List the available probes and select a probe spec for a run. First action: - Install with python -m pip install -U garak, or use a Conda environment and an editable install from source. - Run garak --list_probes to see the probes. - Run garak with a target type, target name and probe spec as in the README examples. Prerequisites and limits: The README says garak is a command-line tool developed in Linux and OSX, and its Conda example uses Python 3.11 or later up to 3.14. Hosted targets need their own API access. - The README is an index to the user guide, which was not read. - Probe results depend on the target and probe set chosen. - Hosted targets may incur provider charges. Source-documented: the README ends with an Apache licence v2 statement. Scan only systems you own or are authorised to test, and follow each provider's terms. Written from the README text only. Statements are as the README gives them; nothing was installed or run. Based on README text only; the user guide, probe lists and code were not read. No install or scan was made. Source inspected: https://raw.githubusercontent.com/NVIDIA/garak/892779207cbc38f65a738ef9cbdfbc1e83aa0b55/README.md ; 20164 bytes ; sha256 640626790dc73543d58857ea0ebc7e47ccdb38cd9dd8fc56467147f7f3a5798e ; 9 exact pinned excerpts in the draft pin map

Structured fields

{
  "revision": {
    "ref": "commit 892779207cbc38f65a738ef9cbdfbc1e83aa0b55",
    "pinned": false,
    "source_url": "https://raw.githubusercontent.com/NVIDIA/garak/892779207cbc38f65a738ef9cbdfbc1e83aa0b55/README.md"
  },
  "checked_at": "2026-11-07T10:50:49.120Z",
  "usefulness": "- Probe a model or chatbot you are authorised to test for hallucination, data leakage, prompt injection or toxicity.\n- Scan models from Hugging Face, OpenAI, Bedrock, LiteLLM, REST endpoints or GGUF files.\n- List the available probes and select a probe spec for a run.",
  "source_type": "repository",
  "availability": "available",
  "source_owner": "garak maintainers (declared repository namespace)",
  "known_limitations": [
    "The README says garak is a command-line tool developed in Linux and OSX, and its Conda example uses Python 3.11 or later up to 3.14. Hosted targets need their own API access.\n\n- The README is an index to the user guide, which was not read.\n- Probe results depend on the target and probe set chosen.\n- Hosted targets may incur provider charges.\n\nSource-documented: the README ends with an Apache licence v2 statement. Scan only systems you own or are authorised to test, and follow each provider's terms.\n\nWritten from the README text only. Statements are as the README gives them; nothing was installed or run.\nBased on README text only; the user guide, probe lists and code were not read. No install or scan was made.",
    "Rendered README text of the exact archived capture at the pinned commit, read in full by the author; no media viewed or heard; nothing executed or installed.",
    "Source-documented steps; execution not performed. Internal affiliated review is not independent validation."
  ]
}

Rights

component	license	hosting	redistribution	modification	training	Evidence and notes
documentation	unknown	permitted	permitted	permitted	unknown	Only the team-authored original attributed summary and links are covered. Upstream papers, tutorials, code, models, weights and media retain their separate terms; no source-text/media redistribution right is implied.

Machine-readable: JSON · Markdown · revisions · changes