EU AI Act enforcement: August 2, 2026

Your AI agents need
tamper-proof audit trails

Open-source compliance infrastructure for LangChain, CrewAI, AutoGen, and OpenAI Agents. Drop-in trust layers that make your agents EU AI Act ready.

Find your compliance gaps in 30 seconds

$ pip install air-compliance && air-compliance scan ./my-project

Zero dependencies. Works on any Python AI project.

Star on GitHub Book a Demo

---

Days

Hours

Minutes

Seconds

air-compliance scan ./my-agent

EU AI Act Compliance Report

Project: ./my-agent | Framework: LangChain

────────────────────────────────────────────

PASS Article 9 — Risk Management ........... 4/4

PASS Article 10 — Data Governance .......... 3/3

PASS Article 11 — Technical Documentation .. 3/3

FAIL Article 12 — Record-Keeping ........... 1/4

WARN Article 14 — Human Oversight .......... 2/4

FAIL Article 15 — Robustness & Security .... 1/4

────────────────────────────────────────────

Coverage: 63% | 14/22 checks passing

CRITICAL: HMAC-SHA256 chain not configured (Article 12)

Everything agents need to comply

Six security and compliance controls that map directly to EU AI Act articles.

🔒

Tamper-Evident Audit Chain

HMAC-SHA256 signed, cryptographically chained logs. Every agent decision is recorded in a chain regulators can mathematically verify.

Article 12

🛡

PII Tokenization

14 built-in detection patterns automatically redact API keys, SSNs, credit cards, and emails before they reach the LLM.

Article 10

⚠️

Consent Gate

Risk-classifies every tool call. Blocks critical operations until approved. Humans stay in control of what the agent can do.

Article 14

🚨

Injection Detection

15+ weighted patterns scan every prompt for injection attacks, jailbreaks, role overrides, and data exfiltration attempts.

Article 15

📚

RAG Write Gate

Source allowlists, content pattern blocking, and rate limits protect your knowledge base from poisoning attacks.

Article 15

📈

Drift Detection

Real-time monitoring for retrieval anomalies: new untrusted sources, trust level shifts, volume spikes, and document dominance.

Article 15

Every major framework. One pip install.

Drop-in trust layers that hook into your existing agent code with 3 lines of setup.

LangChain

LangGraph

pip install air-langchain-trust

CrewAI

Multi-agent

pip install air-crewai-trust

OpenAI Agents

Agents SDK

pip install air-openai-agents-trust

AutoGen

AG2

pip install air-autogen-trust

RAG Pipelines

Knowledge Bases

pip install air-rag-trust

TypeScript

Node.js

npm i openclaw-air-trust

Compliance

Scanner

pip install air-compliance

Gateway

Any HTTP agent

docker pull ghcr.io/airblackbox/gateway

3 lines to comply

Add tamper-evident auditing, PII protection, and injection defense to any framework.

                
                
                
                
            

from air_langchain_trust import AirTrustCallbackHandler

# One handler. Full compliance.
handler = AirTrustCallbackHandler()

# Every LLM call, tool use, and chain step gets
# logged to a tamper-evident HMAC-SHA256 chain.
result = agent.invoke(
    {"input": query},
    config={"callbacks": [handler]}
)

# Auditors verify the chain hasn't been tampered with
handler.verify_chain()  # True
                

from air_crewai_trust import AirTrustHook, AirTrustConfig

# Configure trust controls
hook = AirTrustHook(config=AirTrustConfig())

# Attach to any crew — all agent actions
# are now audited and protected.
crew = Crew(
    agents=[researcher, writer],
    hooks=[hook]
)

crew.kickoff()
                

from air_openai_agents_trust import activate_trust

# One call. Patches the SDK globally.
activate_trust()

# Every agent run is now protected with
# audit trails, PII tokenization, and
# injection detection.
result = await Runner.run(
    agent,
    "Analyze quarterly revenue data"
)
                

from air_rag_trust import AirRagTrust, WritePolicy

# Gate what enters your knowledge base
rag = AirRagTrust(write_policy=WritePolicy(
    allowed_sources=["internal://*"],
    blocked_content_patterns=[r"ignore previous"],
))

# Every document gets SHA-256 provenance
rag.ingest(content=doc, source="internal://kb")
                

Mapped to the regulation

Every control maps directly to a specific EU AI Act article. No guesswork.

Article 9

Risk Management

ConsentGate classifies tool calls by risk level (LOW → CRITICAL) and enforces blocking policies.

Article 10

Data Governance

DataVault tokenizes PII. ProvenanceTracker hashes KB documents with SHA-256 provenance chains.

Article 11

Technical Documentation

Structured audit logging captures full call graphs: chain → LLM → tool → result.

Article 12

Record-Keeping

HMAC-SHA256 tamper-evident chains. Every entry is cryptographically signed and linked to the previous one.

Article 14

Human Oversight

Exception-based blocking interrupts agent execution. Full audit trails enable post-hoc review.

Article 15

Robustness & Security

InjectionDetector, WriteGate, and DriftDetector provide defense-in-depth against adversarial attacks.

Your AI agents needtamper-proof audit trails

Everything agents need to comply

Tamper-Evident Audit Chain

PII Tokenization

Consent Gate

Injection Detection

RAG Write Gate

Drift Detection

Every major framework. One pip install.

3 lines to comply

Mapped to the regulation

5 months left

Your AI agents need
tamper-proof audit trails