Claude, ChatGPT or anything else: when an AI assistant connects to NetSuite from outside, it sees what its role sees. These five checks take an afternoon and save you a bad surprise.
1. Give it its own role
Do not connect an assistant under an administrator or under a real person’s login. Make a role just for it, name it clearly, and start with the least access that still lets it do the job.
2. Read the list of records it can see
Open the role and read the permissions line by line. Ask one question of each: would I paste this into an email to someone outside the company? Item costs, customer pricing, vendor terms and employee records are the usual surprises.
3. Decide on write access
Reading a sales order and changing one are very different risks. Most teams should start read-only, let people get used to the drafts, and add write access one record type at a time.
4. Decide who is allowed to connect
If anyone in the company can link their own AI account to NetSuite, you will not know how many connections exist. Pick the people, write the list down, and review it when someone leaves.
5. Agree how the work gets reviewed
An assistant drafts. A person approves. Decide who that person is for quotes, for supplier messages and for anything that touches money, and decide it before the first draft goes out.
See it instead of guessing
NetSuite AI Monitor shows everything AI can touch in your account, including roles and integrations, so you can check the list above against what is really there. It runs inside NetSuite and the free plan costs nothing.
This post was prompted by Claude for Manufacturing, which launched with a NetSuite connector. The checklist applies to any outside assistant.
This is general guidance, not a security audit. NetIntel is an independent developer and is not affiliated with Anthropic, OpenAI or Oracle.
Kye Gosnell, Founder · Knox Holton, Co-founder and CMO