air-crewai-trust

EU AI Act compliance infrastructure for CrewAI agents. Drop-in trust layer that adds tamper-evident audit logging, PII tokenization, consent-based tool gating, and prompt injection detection - making your CrewAI agent stack compliant with Articles 9, 10, 11, 12, 14, and 15 of the EU AI Act.

Part of the AIR Blackbox ecosystem - the compliance layer for autonomous AI agents.

EU AI Act high-risk obligations take effect December 2, 2027 (deferred from August 2026 by the EU's 2026 omnibus agreement). Regulators expect preparation to be underway now. See the full compliance mapping for article-by-article coverage.

View Interactive Demo: Walk through every feature with animated examples.

Quick Start

pip install air-crewai-trust

from crewai import Agent, Task, Crew
from air_crewai_trust import activate_trust

# One line activates all trust hooks
plugin = activate_trust()

# Your CrewAI code runs as normal - trust layer works transparently
agent = Agent(role="Researcher", goal="Find information", backstory="...")
task = Task(description="Research AI safety", agent=agent)
crew = Crew(agents=[agent], tasks=[task])
crew.kickoff()

# Check what happened
print(plugin.get_audit_stats())
print(plugin.verify_chain())

What It Does

Tamper-Proof Audit Trail

Every tool call and LLM interaction is logged to an HMAC-SHA256 signed chain. Each entry references the previous entry's hash - modify any record and the chain breaks. Like a blockchain, but for your AI agent's actions.

Sensitive Data Tokenization

API keys, credentials, PII (emails, SSNs, phone numbers, credit cards) are automatically detected and replaced with opaque tokens before they reach the LLM. Original values are stored locally in the vault and can be restored when the tool actually needs them.

14 built-in patterns covering:

• API keys (OpenAI, Anthropic, AWS, GitHub, Stripe)
• Credentials (Bearer tokens, private keys, connection strings, passwords)
• PII (emails, phone numbers, SSNs, credit cards)

Consent Gate

Destructive tools (exec, shell, deploy, file_delete) are blocked until the user explicitly approves them. Risk classification:

Prompt Injection Detection

15+ weighted patterns detect prompt injection attempts including role overrides, jailbreaks, delimiter injection, privilege escalation, and data exfiltration. Configurable sensitivity (low/medium/high) and block threshold.

Configuration

from air_crewai_trust import activate_trust, AirTrustConfig

config = AirTrustConfig(
    consent_gate={
        "enabled": True,
        "always_require": ["exec", "spawn", "shell", "deploy"],
        "risk_threshold": "high",
    },
    vault={
        "enabled": True,
        "categories": ["api_key", "credential", "pii"],
    },
    injection_detection={
        "enabled": True,
        "sensitivity": "medium",
        "block_threshold": 0.8,
    },
    audit_ledger={
        "enabled": True,
        "max_entries": 10000,
    },
    # Optional: forward audit records to AIR Blackbox gateway
    gateway_url="https://your-gateway.example.com",
    gateway_key="your-api-key",
)

plugin = activate_trust(config)

CrewAI Hook Mapping

API Reference

from air_crewai_trust import activate_trust, deactivate_trust, get_plugin

# Activate / deactivate
plugin = activate_trust(config=None)  # Returns AirTrustPlugin
deactivate_trust()                     # Unregisters all hooks

# Plugin methods
plugin.get_audit_stats()   # → {"total_entries": 42, "chain_valid": True, ...}
plugin.verify_chain()      # → {"valid": True, "total_entries": 42}
plugin.export_audit()      # → [{"id": "...", "action": "tool_call", ...}, ...]
plugin.get_vault_stats()   # → {"total_tokens": 5, "by_category": {"api_key": 3, "pii": 2}}

EU AI Act Compliance

See docs/eu-ai-act-compliance.md for the full article-by-article mapping.

AIR Blackbox Ecosystem

Development

git clone https://github.com/airblackbox/trust-crewai.git
cd trust-crewai
pip install -e ".[dev]"
pytest tests/ -v

License

Apache-2.0