Contact Us
A netsuite consultant seated at a desk using a laptop, coffee cup in workspace.

NetSuite Applications Suite - Associated Risks, Controls, and Mitigation Strategies

There are inherent risks in using an AI client to extract, process, or update NetSuite data via the AI connector and MCP tools. This article outlines three main risks, the controls NetSuite already uses, and the actions users should take to mitigate risks.

1. User Access

Unauthorized or excessive access occurs when an AI agent operates beyond the user's intended permissions — invoking elevated scripts, accessing admin features, or acting outside the scope of the authorizing user's role.

NetSuite Controls

This is the strongest layer of defense. NetSuite enforces the following regardless of what the AI client attempts:

  • MCP tools run within the authorizing user's role and permissions — the agent cannot do anything you cannot do (unless you have REST Web Services permissions set to ‘Full’ for the role - see additional mitigations).
  • Administrator roles and roles with full permissions cannot be used with MCP.
  • "Run as role" is disabled; tools cannot invoke scripts with elevated privileges.
  • Tools cannot invoke Suitelets or make API requests to external endpoints.
  • OAuth 2.0 authorization is required per user, per agent.
  • All MCP tool usage is logged in NetSuite.

Additional Mitigations

  • Grant MCP permission only to users who need it.
  • Remove REST Web Services permissions from any roles with MCP connection if record creation or update is unnecessary. This permission can extend the AI client’s capabilities past the normal role permissions given to the user via the UI.
  • Avoid granting MCP to high-privilege roles.
  • Create dedicated, narrowly scoped roles for MCP use.
  • Use MCP tool namespaces to limit which tools are available in a given session.
  • Review MCP permissions regularly.

2. Hallucination

Hallucinations occur when the model generates information that appears accurate but is fabricated or incorrect. In an MCP context, this can mean acting on a wrong figure, or the agent invoking a tool — making a payment, granting an approval, modifying a record — without the user explicitly asking for it.

NetSuite Controls

When using MCP tools to retrieve data, the AI client is grounded in live NetSuite records and reports via ns_getRecord, ns_runReport, or ns_runSearch. The values it returns should reflect actual system data. Invoking ns_runCustomSuiteQL generates an analysis of custom query results generated by the AI client - these results may not always be accurate. The majority of hallucination risk lives downstream of the tool call. The agent might misrepresent a large result set, perform incorrect math on top of returned figures, or infer you want it to take an action when you only meant to ask a question.

  • All MCP tool invocations are logged, giving you a full audit trail of every action the agent took on your behalf.

See NetSuite AI Connector Service-specific Integration Record and NetSuite AI Connector Service (MCP) Execution Log.

  • Before calling ns_runCustomSuiteQL, the AI client calls ns_getSuiteQLMetadata to identify all related records in the NetSuite environment.

User Mitigation Strategies

  • Verify agent outputs against actual NetSuite records or reports before taking action.
  • Ask the AI client to return the SuiteQL queries used.
  • Start with a narrow scope when testing new agents or tools.
  • Review audit logs regularly.

See NetSuite AI Connector Service-specific Integration Record and NetSuite AI Connector Service (MCP) Execution Log.

3. Prompt Injection

Prompt injection occurs when hidden instructions embedded in content — pasted code, copied web text, third-party agent prompts, or even a NetSuite record field — redirect the AI client to perform unintended actions.

NetSuite Controls

NetSuite limits the blast radius of a successful injection by removing outbound paths from the MCP toolset. Tools cannot perform HTTP requests to external destinations and cannot invoke SuiteScripts with external API calls, so data retrieved into the context window has nowhere to go unless you have a connected email or file system tool active in the same session.

Additional Mitigations

  • Remove REST Web Services permissions from any roles with MCP connection to prevent the AI client from updating or creating records. This permission can extend the AI client’s capabilities past the normal role permissions given to the user via the UI.
  • Review any large block of text before submitting it to your AI client.
  • Vet third-party agents and MCP servers before connecting them.
  • Enable only the MCP tools your session actually needs.
  • Prefer agents that ask for confirmation before executing sensitive actions.
  • Be especially cautious when running email or file system connectors alongside NetSuite MCP tools.

Related Articles

Tagged with AI, Security & Permissions