Free AI tools and your data: what "we don't train on your data" really protects
For large customers, "no training on your data" is usually a contractual commitment. Below the enterprise threshold it is often an account setting, which the vendor can change in a release note. Small businesses using free or self-serve AI tools should pick the most protective settings, assess the remaining risk to personal data, and document that assessment.
Key takeaways
- A contractual no-training commitment and a no-training setting are not the same protection.
- Free tiers are funded by something. Find the sentence that says what.
- One AI lab's own terms describe its free access as a company commitment, not a technical guarantee.
- A privacy standard that assumes a negotiated contract leaves one-person companies with invisible non-compliance.
- The practical posture: protective settings, a written risk assessment, and a review when terms change.
Why do two customers of the same AI tool get different protection?
I made a submission to the PDPC's consultation on personal data in generative AI as a deployer, meaning a company that builds on a model it doesn't train. One point I raised has aged fast.
A multinational and a one-person company use the same AI tool. The multinational's contract says "we won't train on your data." The one-person company gets a checkbox. Those look like the same protection. They aren't.
What does a free tier change?
- Enterprise terms have a size threshold. Below it, "no training" is an account setting, not a contractual commitment. A setting can change in a release note. A signed contract can't.
- Free tiers are funded by something. Sometimes venture capital, honestly disclosed. Sometimes your inputs. The question is whether you can point to the sentence that tells you which.
- The commitment isn't a guarantee. Agnes AI's own terms describe its free access as a company commitment, not a technical guarantee, and tell developers to account for pricing changes before building production dependencies. It is one of the most useful sentences in AI, and it sits in a developer document almost nobody reads.
Why does this matter for one-person companies?
AI is driving the rise of the one-person company. More businesses now operate below the threshold where any vendor will negotiate with them. A privacy standard that assumes a negotiated contract pushes everyone below that line toward the same invisible non-compliance as shadow AI.
What can a small business do today?
- Choose the most protective settings available. Turn off training and history where you can.
- Read the terms for three things. Training use, retention period, and who else can access your data.
- Keep personal data out where possible. Remove names and identifiers before you paste.
- Assess what is left and write it down. A short note of the tool, the data, the settings and the remaining risk.
- Review when the terms change. Treat a pricing or terms update as a trigger to reassess.
This isn't a workaround. For many small businesses it is the whole compliance posture available. Doing it deliberately is the difference between a gap you have managed and a gap you will discover.
Frequently asked questions
Can a small business use free AI tools under the PDPA?
It can, but it remains responsible for personal data it puts into those tools. That means understanding how the provider uses the data, choosing protective settings, and making reasonable security arrangements. This is general information, not legal advice.
What is the difference between a no-training setting and a no-training clause?
A clause is a contractual promise the vendor can't change on its own. A setting is a product feature that can change in an update, often with little notice.
What should a small business document when using AI tools?
Which tools are used, what data goes in, which settings are chosen, what the terms say about training and retention, the remaining risk, and when the assessment will be reviewed.
This article is general information, not legal advice. It reflects the position as at the date of publication. A plain-text version for AI assistants is at /blog/free-ai-tools-data-protection-small-business.md.