Security Best Practices

Guidelines for securing your Agent Action Firewall deployment and protecting your AI agents.

API Key Management

Key Rotation

Rotate API keys regularly and immediately if compromised:

1. Generate a new key in Settings → API Keys
2. Update your agent configuration with the new key
3. Verify the agent works with the new key
4. Revoke the old key

Key Storage

• Never commit API keys to version control
• Use environment variables or secrets managers (Vault, AWS Secrets Manager)
• Restrict key access to only necessary services
• Use separate keys for development, staging, and production

Key Scoping

Create dedicated keys for each agent with minimal permissions:

// Good: Dedicated key per agent
const orderAgent = new AgentFirewallClient({
  apiKey: process.env.ORDER_AGENT_KEY,
  agentId: 'order-processor',
});

const supportAgent = new AgentFirewallClient({
  apiKey: process.env.SUPPORT_AGENT_KEY,
  agentId: 'support-bot',
});

Policy Configuration

Default Deny

Start with a deny-all policy and explicitly allow needed operations:

package aaf.policies

default decision = {"action": "deny", "reason": "No matching allow rule"}

# Explicitly allow specific operations
decision = {"action": "allow"} {
  input.tool == "http_proxy"
  input.operation == "GET"
  allowed_domains[input.params.url]
}

allowed_domains[url] {
  startswith(url, "https://api.trusted.com/")
}

Risk-Based Approvals

Require human approval for high-risk operations:

decision = {
  "action": "require_approval",
  "risk_level": "high",
  "reason": "Write operations require approval"
} {
  input.operation in ["POST", "PUT", "DELETE"]
  not input.context.pre_approved
}

Rate Limiting

Limit action frequency to prevent abuse:

decision = {
  "action": "deny",
  "reason": "Rate limit exceeded"
} {
  rate_limit_exceeded(input.agent_id)
}

Network Security

SSRF Protection

The HTTP proxy connector automatically blocks:

• Private IP ranges (10.x.x.x, 172.16-31.x.x, 192.168.x.x)
• Localhost (127.0.0.1, ::1)
• Cloud metadata endpoints (169.254.169.254)
• Internal hostnames

Domain Allowlisting

Restrict outbound requests to approved domains:

// In your policy or connector config
const allowedDomains = [
  'api.stripe.com',
  'api.github.com',
  'slack.com',
];

TLS Requirements

• Always use HTTPS for API communication
• Enforce TLS 1.2+ for all connections
• Validate SSL certificates (no self-signed in production)

Data Protection

Sensitive Data Handling

• Redaction: Automatically redact sensitive fields from logs
• Encryption: Sensitive params encrypted at rest
• Retention: Configure audit log retention policies

Configuring Redaction

// Redact sensitive fields before logging
const sensitiveFields = [
  'password',
  'api_key',
  'token',
  'credit_card',
  'ssn',
];

Audit Trail Integrity

Audit events use cryptographic hash chaining for tamper evidence. Each event includes a hash of the previous event, creating an immutable chain.

Access Control

Role-Based Access

Organization Isolation

Multi-tenant isolation is enforced at the database level with Row-Level Security (RLS). Each organization can only access its own data.

Monitoring and Alerting

Security Events to Monitor

• Failed authentication attempts
• Policy denials (especially repeated)
• Unusual action patterns
• High-risk action submissions
• Approval timeouts

Setting Up Alerts

// Example: Alert on repeated denials
if (denialCount > 10 && timeWindow < '5m') {
  sendAlert({
    severity: 'warning',
    message: 'Unusual denial rate for agent',
    agent_id: agentId,
  });
}

Incident Response

If an API Key is Compromised

1. Immediately revoke the key in Settings → API Keys
2. Review audit logs for unauthorized actions
3. Generate a new key and update your agent
4. Investigate how the key was exposed
5. Document the incident

Exporting Evidence

Use Proof Packs to export forensic evidence:

# Export proof pack for an action
curl -H "Authorization: Bearer $TOKEN" \
  "https://api.agentactionfirewall.com/v1/proof-pack/action-id" \
  -o proof-pack.zip

Compliance

Audit Requirements

• All actions are logged with timestamps
• Hash chaining provides tamper evidence
• Proof packs support compliance exports
• Configurable retention periods

Data Residency

Cloud-hosted data is stored in your selected region with full encryption at rest and in transit.

Security Checklist

Need Help?

For security questions or to report vulnerabilities, contact info@agentactionfirewall.com.