AI Potluck
Product / UX / Safety & Guardrails

LLM Guard

Protect AI

Security toolkit for LLM interactions from Protect AI, providing input and output scanners for prompt injection, PII and data leakage, secrets, toxicity, bias and harmful content, with detection and sanitization. It runs as a Python library on infrastructure you operate, or deployed as an API.

Protect AI's commercial platform is a separate product. Verified 2026-08-13 via GitHub.

Openness

5 high confidence
5.0
license
MIT(OSI)
source
public(full Python implementation)
self-host
yes
service
none(the OSS toolkit
core-gated
ungated

Fully open source: MIT-licensed, complete source, self-hostable. (Protect AI's commercial platform is a separate product; this is the standalone OSS toolkit.)

  • https://github.com/protectai/llm-guard recorded 2026-08-13

    Repo metadata records license spdxId - MIT. README calls it 'an open source solution', documents installing and customizing it yourself and deploying it as an API, and enumerates the prompt and output scanners in the published package; there is no enterprise directory, commercial edition or license key on the page.

Adoption

3 high confidence
3.0

212,987 downloads in the trailing 30 days for the PyPI package `llm-guard`, which bands at 100K-1M, level 3. The package is the product's own rather than a same-named stranger - its PyPI project URLs point at https://github.com/protectai/llm-guard - and for a Python library PyPI is the primary distribution channel rather than a minority one, so downloads rather than GitHub stars are what the level rests on.

Capability

3 medium confidence
3.0

A solid set of input and output scanners, lighter-weight than NeMo Guardrails but easy to drop in: a scanner library with no policy language and no dialog control sits one rung below NeMo Guardrails.

  • https://github.com/protectai/llm-guard recorded 2026-08-13

    README lists the prompt scanners (Anonymize, BanCode, BanCompetitors, BanSubstrings, BanTopics, Code, Gibberish, InvisibleText, Language, PromptInjection, Regex, Secrets, Sentiment, TokenLimit, Toxicity) and the output scanners (BanCode, BanCompetitors, BanSubstrings, BanTopics, Bias, Code, Deanonymize, JSON, ...), and frames the toolkit as sanitization, detection of harmful language, prevention of data leakage and resistance to prompt injection.

Verified 2026-08-13