Is It Safe to Put Client Data Into ChatGPT?
Staff are already using AI tools on real work, whether or not there's a policy. The risk isn't the technology — it's which account the data goes through. Here's what actually differs, and how to fix it.
By the time leadership asks whether AI is safe to use, staff have usually been using it for months. Not maliciously — because it genuinely saves time.
So the useful question isn’t “should we allow AI?” The material is already going somewhere. The useful question is which account it’s going through, and under what terms.
The distinction that matters: which tier
The same tool behaves very differently depending on how you access it.
Consumer accounts — the free or individual paid tiers people sign up for with a personal email — may use conversations to improve models unless the user switches that off. Retention and controls are set by the provider, not by you.
Business, enterprise and API tiers are generally excluded from model training by default, come with defined retention, and give an administrator control over who has access. They’re also covered by an agreement between your business and the provider.
Terms change. Always check the current data-usage policy for the specific plan you’re on.
The practical implication: an employee pasting a client contract into a personal account and an employee pasting it into a business account under your agreement are two very different events, even though the screen looks identical.
A simple way to classify what goes where
Routine work — drafts, general research, rewriting your own marketing copy — is usually fine on a business or API tier.
Client-confidential material — contracts, personal information, anything covered by a confidentiality clause — needs a deliberate decision. Either a business tier whose terms you’ve actually checked, or a model that runs on your own infrastructure.
The most sensitive material — privileged documents, health data — is where self-hosted models earn their keep. The data never leaves your network, which removes the disclosure question entirely.
What a safe setup looks like
- Find out what’s already happening. Which tools, which accounts, what kind of material. It’s almost always more than leadership expects.
- Offer a sanctioned option that’s better than the workaround. People stop using personal accounts when the company account is just as good.
- Write a one-page policy. What can go where. Short enough that people actually read it.
- Log usage on the business account, so there’s a record if a question ever comes up.
Why a ban makes it worse
Prohibition doesn’t stop usage; it relocates it. People move to personal accounts and personal phones, where you have no contract, no logs and no idea what’s been shared. A sanctioned path with clear rules is far safer.
This is exactly the situation in staff are already pasting client data into ChatGPT. The wider work lives in our AI and automation practice.
- AI
- Data Privacy
- ChatGPT