crimsonguard 0.1.0

Trust and assurance tool for AI — apply NIST AI RMF and ISO/IEC 42001 to LLMs and MCP servers.

CrimsonGuard

Trust your AI. Prove it.

CrimsonGuard applies NIST AI RMF 1.0 and ISO/IEC 42001:2023 to make AI systems reliable, safe, and secure. It produces auditor-ready evidence (OSCAL assessment-results + PDF) mapped to standard controls.

Two audit modes

• LLM mode — adversarial probes against LLM-powered AI: hallucination, prompt injection, PII leak, bias, and more.
• MCP mode — security audit against MCP tool servers: indirect prompt injection via tool output, input validation, authz boundary, audit log completeness.

Every finding is tagged to NIST AI RMF subcategories, ISO/IEC 42001 Annex A controls, OWASP GenAI Top 10, and MITRE ATLAS techniques — so you get technical findings and compliance evidence in one scan.

Status

Early development (v0.0.1, alpha). Public v0.1 target: ~6 weeks.

Install (once released)

pip install crimsonguard

Quick start (planned)

# Audit an MCP server
crimsonguard scan-mcp --target bfd-gen --profile process-safety

# Audit an LLM endpoint
crimsonguard scan-llm --target https://api.openai.com --model gpt-4o-mini

# Generate OSCAL + PDF report
crimsonguard report --scan-id <id> --format oscal,pdf

License

Apache-2.0. See LICENSE.

About

Built by the-s-lab — trustworthy AI for process engineering and safety-critical industries.