air-blackbox 1.9.0

AI governance control plane — compliance, inventory, incident response, and audit for AI agents

AIR Blackbox

The EU AI Act enforcement deadline is August 2, 2026. Scan your AI agent in 5 minutes. Get audit-ready evidence for four compliance frameworks from a single scan.

Install

pip install air-blackbox

First scan in 30 seconds

air-blackbox comply --scan .

That's it. 39 checks across 6 EU AI Act articles. No config, no API keys, no Docker.

Want the demo instead? Generate sample data and see everything in action:

air-blackbox demo

What you get

Article  9 (Risk Management):      PASS  (5 checks)
Article 10 (Data Governance):      WARN  (2 findings)
Article 11 (Technical Docs):       FAIL  (3 findings)
Article 12 (Record-Keeping):       PASS  (4 checks)
Article 14 (Human Oversight):      WARN  (1 finding)
Article 15 (Robustness):           PASS  (3 checks)

Compliance Summary: 22 passing  3 warnings  3 failing  out of 28 checks

Run with -v to see fix hints for each failing check.

Every check tells you what's wrong and how to fix it. Run with -v for remediation guidance.

One scan, four compliance standards

Most compliance tools cover one framework. AIR Blackbox maps every check to four simultaneously:

air-blackbox standards

Category	EU AI Act	ISO 42001	NIST AI RMF	Colorado SB 205
Risk Management	Article 9	6.1, 6.1.2, A.6.2.1	GOVERN 1, MAP 1, MAP 3	Section 6(2)(a-b)
Data Governance	Article 10	A.6.2.4, A.6.2.5	MAP 2, MEASURE 2	Section 6(2)(c)
Documentation	Article 11	7.5, A.6.2.2	GOVERN 4, MAP 5	Section 6(2)(d), 6(3)
Record-Keeping	Article 12	A.6.2.6, 9.1	MEASURE 1, MANAGE 4	Section 6(4)
Human Oversight	Article 14	A.6.2.3	GOVERN 2, MANAGE 1	Section 6(2)(e), 7
Robustness	Article 15	A.6.2.8, A.6.2.9	MEASURE 3, MANAGE 2	Section 6(2)(f)
Consent	GDPR Art. 6/7	A.6.2.5, A.6.2.11	GOVERN 3	Section 5
Bias/Fairness	Article 10	A.6.2.4, A.6.2.10	MAP 2, MEASURE 2	Section 2, 6(2)(c)

Filter by framework:

air-blackbox comply --frameworks eu,iso42001       # Just EU + ISO
air-blackbox comply --frameworks colorado           # Just Colorado
air-blackbox standards -f nist                      # NIST detail view
air-blackbox standards --lookup "Article 9"         # Reverse lookup

Choose your path

I want a quick compliance scan -- Use the CLI (you're here)

air-blackbox comply --scan . -v

I want runtime audit trails -- Add air-trust to your AI client

pip install air-trust

import air_trust
client = air_trust.trust(OpenAI())  # every call is now HMAC-SHA256 audited

I want human approval gates -- Add air-gate for tool-level oversight

pip install air-gate

I need auditor-ready evidence -- Export a signed bundle

air-blackbox export --format pdf

I use Claude Desktop or Cursor -- Install the MCP server

pip install air-blackbox-mcp

All commands

Command	What it does
air-blackbox comply	EU AI Act compliance scan with multi-framework crosswalk
air-blackbox standards	Browse the 4-framework standards crosswalk
air-blackbox discover	Shadow AI inventory + CycloneDX AI-BOM generation
air-blackbox replay	Incident reconstruction from HMAC audit chain
air-blackbox export	Signed evidence bundle (JSON or PDF) for auditors
air-blackbox validate	Pre-execution runtime checks for agent actions
air-blackbox history	Compliance score trends over time with diff
air-blackbox demo	Zero-config demo with sample data
air-blackbox test	End-to-end stack validation
air-blackbox setup	One-command model install + verification
air-blackbox init	Create compliance doc templates in your project

Framework trust layers

Drop-in compliance for your existing AI stack. No code changes beyond the import:

pip install air-blackbox[langchain]    # LangChain / LangGraph
pip install air-blackbox[crewai]       # CrewAI
pip install air-blackbox[openai]       # OpenAI Agents SDK
pip install air-blackbox[all]          # Everything

Auto-detected frameworks: OpenAI, Anthropic, Google ADK, LangChain, CrewAI, LlamaIndex, Haystack, AutoGen, Semantic Kernel, Smolagents, PydanticAI, DSPy, Browser Use, and more.

How it works

1. Static analysis -- Scans your Python code for compliance patterns (error handling, logging, PII detection, human oversight gates, documentation)
2. Hybrid AI analysis -- Optionally uses a local LLM (air-compliance model via Ollama) for deeper assessment beyond regex
3. Runtime checks -- With trust layers installed, validates live AI traffic against EU AI Act requirements
4. Evidence export -- Packages compliance results + AI-BOM + audit trail into a single verifiable document

Your code never leaves your machine. No cloud, no API keys, no vendor lock-in.

Why this exists

The EU AI Act (Regulation 2024/1689) requires organizations deploying high-risk AI systems to demonstrate compliance across Articles 9-15. Penalties reach 35M euros or 7% of global turnover. The first enforcement deadline is August 2, 2026.

Existing compliance tools are either manual checklists that take weeks, enterprise SaaS that costs thousands per month, or generic linters that don't understand AI-specific requirements.

AIR Blackbox is a developer tool that runs locally, scans in minutes, and produces audit-ready evidence. Open source, no vendor lock-in, and free.

Links

• Website: airblackbox.ai
• Interactive demo: airblackbox.ai/demo/signed-handoff
• Repository: github.com/airblackbox/gateway
• MCP server: github.com/airblackbox/air-blackbox-mcp
• Changelog: CHANGELOG.md
• PyPI: pypi.org/project/air-blackbox

License

Apache-2.0. See LICENSE.