guardrailprobe

Provider-agnostic AI guardrail benchmarking.

Tests your guardrail layer — not your model — with 78 probes across the OWASP LLM Top 10 and content-moderation categories, against 10 guardrail backends. Deterministic scores, signed reports.

• 78 adversarial probes, OWASP LLM Top 10 coverage
• 10 backends: NeMo, Lakera, Presidio, Azure, Bedrock & more
• JSON / Markdown / signed PDF reports

pip install guardrailprobe
guardrailprobe run --backends lakera

The source repository opens to the public in late July 2026 — the package is installable from PyPI today.