Enabling External AI Agents with Security Controls in NetSuite

Enabling external AI agents in NetSuite allows organizations to leverage advanced AI capabilities, enhancing operational efficiency. However, their integration introduces security concerns that need careful management.

Risks of Using External AI Agents

External AI agents and large language models (LLMs) come with certain risks, including:

• Prompt Injection: Malicious actors may embed instructions in content processed by LLMs, leading to unauthorized actions or data leaks.
• Hallucination: LLMs can generate incorrect or fabricated information, posing risks to decision-making processes.

These risks can result in:

• Unintended Actions: AI agents might execute transactions or command without user intent.
• Data Corruption: Actions may inadvertently modify or delete data.
• Sensitive Information Disclosure: Potential exposure of confidential data to unauthorized users.

Controls Available in NetSuite

While neither prompt injection nor hallucinations can be fully mitigated by NetSuite, several controls can help minimize risks:

• User Permissions: Administrators control who has access to Managed Cloud Platform (MCP) tools, with no permissions granted by default.
• Limited API Access: MCP tools work with user permissions but cannot perform certain high-risk actions, such as invoking Suitelets or external HTTP requests.
• Logging and Tracking: All MCP tool usage is logged for accountability.
• Explicit Consent: Users must authorize the AI agent during the OAuth 2.0 authorization process.
• Tool Scoping: Administrators can limit the tools available to AI agents, enhancing security.

Steps to Enable External AI Agents

To enable external AI agents, follow these coordinated steps:

For Account Administrators

1. Assign MCP Permissions: Grant permissions to users who require access.
2. Install MCP Tools: Setup the MCP tools defining allowed actions for AI agents.

Important: The AI actions are limited to those defined by installed tools, with user permissions determining functionality.

For End Users

1. Configure the AI Agent: Set up the external AI agent according to your requirements.
2. Authorize the Agent: Ensure the AI agent is permitted to act on your behalf within NetSuite.

Mitigation Strategies

To combat risks effectively, consider the following strategies:

• Vendor and Tool Trustworthiness: Always use reputable AI agents and tools. Understand their security measures regarding prompt injection and hallucination.
• Access Management: Limit MCP permissions to only necessary personnel and roles, avoiding high-privilege users for MCP access.
• Scope Limitation: Only activate essential MCP tools to minimize exposure.
• User Awareness: Train users on risks and encourage caution during ML interactions.
• Technical Safeguards: Use secure environments when employing MCP tools affecting sensitive systems.

Compliance Risks

Be aware of compliance limitations that various regulations impose on your use of AI tools, especially in sensitive areas like finance or HR. Different jurisdictions may have varying requirements affecting tool utilization.

Key Takeaways

• External AI agents in NetSuite can enhance operational efficiency but carry inherent risks.
• Strict permissions and tool limitations are critical for reducing potential misuse.
• Continuous user training and awareness are essential for safe AI agent operations.