Insights / Security
Is it safe to put customer data in ChatGPT? A manufacturer's guide
Your team is already pasting quotes, customer lists and drawings into AI chat apps. Here is how to think about the risk, the questions to ask any AI vendor, and a policy you can put in place this month.
Somewhere in your company today, someone pasted a customer email into an AI chat app and asked it to draft a reply. Someone else uploaded a spreadsheet of open quotes and asked for a summary. Maybe a drawing or a process document went in too.
None of this is malicious. AI chat tools are genuinely useful, and your people are trying to work faster. But manufacturers hold some of the most sensitive data in any industry: customer pricing, supplier terms, controlled technical data, and process knowledge that is the business itself. So the question is a fair one. Is it safe?
The honest answer: it depends on which product, which plan, which settings, and what you put in it. This guide breaks that down.
Consumer and enterprise AI are not the same product
The same model can sit behind very different products. The difference that matters is the terms and controls around it.
- Consumer chat apps, including free and individual paid plans, are built for individuals. Depending on the provider, plan and settings, your conversations may be stored and may be used to improve the provider's models. Policies differ between providers and change over time.
- Business and enterprise offerings typically come with stronger commitments: no training on your data by default, admin controls, retention settings and contractual terms.
- API access, the way software products like ours connect to models, usually comes with business terms that are different again from the consumer app.
The practical risk in most companies is not the enterprise tier. It is the personal account an employee signed up for on their own, with default settings and no admin oversight.
What can go wrong
The risks are less dramatic than headlines suggest, but they are real:
- Loss of control. Data in a personal account sits outside your systems. You cannot see it, audit it or delete it, and when the employee leaves, it leaves with them.
- Training and retention. Depending on the product and settings, prompts may be kept and used to improve models. Even when that risk is small, it is not one your customers agreed to.
- Contract and regulatory exposure. Customer agreements often restrict how you share their data. Some technical data is subject to export controls, and some industries carry their own compliance obligations. A copy and paste can quietly break those terms.
- No record of what happened. If something goes wrong, there is no company audit trail of what was shared, by whom, or what the AI did with it.
The risk is rarely the model. It is data leaving the boundary your company controls, with nobody keeping track.
Questions to ask any AI vendor
Whether you are evaluating an enterprise chat plan or an AI platform like ours, these questions separate real protection from marketing:
- Is our data used to train models? Get the answer in writing, for the vendor and for every model provider they route to.
- How long are prompts and outputs kept, and can we delete them?
- How is our data separated from other customers? "Logically separated" can mean many things. Ask where the separation is enforced.
- Who in our company can see what? Look for role-based access that matches how your organization already works.
- Can the AI take actions, and what stops it from taking the wrong one? If it can send email or update records, ask what approval sits in between.
- Is there an audit trail? You should be able to see what was done, by whom, and when.
- Where is the data hosted, and who are the sub-processors?
- What certifications do you hold, or are working toward? SOC 2 is the common baseline for B2B software.
A simple policy you can adopt this month
You do not need a 40-page AI policy to get most of the benefit. Start with three tiers:
- Public information: marketing copy, published specs, general research. Fine for any approved AI tool.
- Internal information: internal processes, non-customer business data. Only in company-approved tools with business terms.
- Restricted information: customer data and pricing, supplier terms, drawings, controlled technical data, personal information. Only in tools approved for it, with access controls and an audit trail, or not at all.
Then make the approved option easy. People turn to personal accounts because they work. If the sanctioned tool is clumsy, the policy will be ignored. Pair the policy with a short training session and a named person who answers "can I use AI for this?" questions.
Where InnateIQ fits
We built InnateIQ for exactly the data this guide is about. In practice that means:
- InnateIQ never uses your data to train models, and OTIS runs on enterprise model APIs, not consumer chat apps.
- Each company's data is isolated by row-level security in the database itself, and denied by default.
- Your roles apply to the AI too. OTIS can only see what the person asking can see.
- Nothing changes without a person's approval. Updating a record or sending an email is shown as a pending decision first.
- It is hosted on AWS in the United States, encrypted at rest, inside a private network.
You can read the details on our security page.
The bottom line: AI chat is not inherently unsafe, but a personal account is the wrong home for customer data. Decide what data goes where, ask vendors hard questions, and give your team an approved tool that is good enough that they actually want to use it.
