opengap

Least-privilege permissions

• Grant only what's needed. Each AI agent should have access to only the specific tools, files, and systems required for its job—not blanket access to your entire network or data.
• Use separate credentials per agent. Create unique login accounts or API keys for each agent rather than sharing one master credential, so you can revoke individual agents without disrupting others.
• Rotate and expire credentials regularly. Set automatic expiration dates for API keys and passwords (e.g., every 90 days) so old compromised credentials become useless over time.
• Require approval for sensitive actions. Configure agents so they cannot modify critical systems, delete data, or transfer money without explicit human sign-off.

Network isolation

• Restrict outbound connections. Prevent agents from reaching unexpected external services; whitelist only the legitimate APIs and databases they need to communicate with.
• Run agents in sandboxed environments. Use isolated containers or virtual machines so an agent cannot escape and access your broader company network or other sensitive systems.
• Segment your network. Place agents in a separate network zone away from financial systems, personnel records, and other high-value targets.
• Block direct internet access when possible. Route all agent communications through a controlled gateway that logs and inspects traffic before it leaves your organization.

Monitoring

• Log all agent actions. Record what data each agent accessed, what instructions it executed, and what results it produced—similar to surveillance camera footage.
• Set up alerts for anomalies. Configure automatic warnings if an agent suddenly makes unusual requests, accesses files it normally ignores, or communicates with unknown external servers.
• Audit git repository changes. Track who modified agent definitions, when, and what changed; investigate unexpected or unauthorized updates.
• Review credential access patterns. Monitor when and how agents use passwords or API keys to spot signs of theft or misuse.

Red flags to watch for

• Secrets in git history. Scan your repository for exposed API keys, passwords, or authentication tokens; if found, revoke them immediately as they may have been copied during backup or transfer.
• Unexpected repository access. If your git repository is accessed by unfamiliar accounts, unusual locations, or at odd hours, treat it as a potential breach.
• Agent behavior changes. An agent suddenly requesting new permissions, accessing different systems, or producing different outputs may indicate unauthorized modification or compromise.
• Fragmented or unsupported implementations. If your vendor stops maintaining their OpenGap implementation or if different vendors' versions become incompatible, you lose visibility and control—migrate to better-maintained standards.