neo

Least-privilege permissions

• Restrict credential scope: Grant Neo only the minimum credentials needed for its specific task (e.g., read-only database access, not admin keys). Avoid sharing credentials across multiple AI models in the swarm.
• Separate service accounts: Create isolated accounts for Neo operations rather than reusing shared team credentials, so compromised tokens affect only that connector.
• Audit credential usage: Regularly review what credentials Neo has requested or used in the past 30 days; revoke any that are no longer needed.
• Rotate tokens frequently: Set automatic expiration of API tokens and refresh keys every 7–30 days to limit window of exposure if leaked.

Network isolation

• Run in sandboxed environment: Deploy Neo in a network segment separate from critical databases, payment systems, and customer data—use firewalls or private VPCs to restrict outbound traffic.
• Whitelist outbound connections: Allow Neo to contact only known, necessary external services (e.g., specific AI model APIs); block everything else by default.
• No direct internet access: If possible, route Neo's requests through a corporate proxy or gateway so you can log and inspect what data leaves your network.
• Air-gap sensitive operations: Keep Neo away from production systems handling personal data, financial records, or health information until you've completed security review.

Monitoring

• Log all AI decisions and changes: Capture what Neo modifies in your applications, which data it accessed, and why—store logs for at least 90 days in a read-only archive.
• Alert on self-modifications: Set up immediate notifications if Neo attempts to change its own code, disable logging, or escalate its permissions.
• Track data flows: Monitor what customer or business data Neo ingests into its GraphRAG system and where it goes; flag any unusual third-party data transfers.
• Human review gates: Require a person to approve Neo's production changes before they execute, rather than letting it auto-heal or self-modify without oversight.

Red flags to watch for

• Opaque self-healing: If Neo modifies its own behavior or recovery logic and you cannot explain why, pause operations and review the change before resuming.
• Unexplained token usage: Sudden spikes in API calls, credential access, or tokens being used by services you didn't authorize—investigate immediately.
• Data retention silence: If Neo's creator cannot clearly describe how long it keeps your application data, where it's stored, and who can access it, treat as high-risk until answered.
• Pre-production status: Neo has very few active users and limited security review; use only in non-critical environments or as a pilot with strong safeguards and exit plan.