AI Usage Policy

How the AI in this product is constrained, what it may receive, and what you may not use it for

2026/09/07

This policy describes how the AI component of this service is constrained, what data it may receive, and how it may be used. It supplements the Terms of Service and Privacy Policy.

What the AI is allowed to do

The AI has exactly one job: read a statistical profile of your file and propose a cleaning plan as a list of structured rules.

  • The AI cannot modify your data. It only produces rule proposals.
  • Every rule must be confirmed by you before anything runs. Execution is performed by a deterministic engine, not by the model.
  • The AI can only emit operations from a fixed allowlist. It cannot output SQL, JavaScript, Python, R or any other executable code that we then run.
  • Your original file is never modified or overwritten. Results are written to new files.
  • AI proposals can vary. Once you confirm a plan, the service freezes its rule snapshot. Running the same input with that snapshot and the same engine version produces a reproducible, auditable result.

What the AI receives

The model never receives your file. It receives only:

  • Column names, inferred types, missing-value counts and unique-value counts.
  • A short list of detected data-quality issues.
  • The dataset context, data dictionary and safeguards you provide to guide the cleaning plan. Avoid entering unnecessary personal or confidential data in these fields.
  • Only if you enable sample sharing, at most five sample rows, with values such as email addresses, phone numbers, long identifiers and URLs masked before they are sent. Sample sharing is off by default.

We do not use your file contents to train models.

Third-party AI provider

TidyX uses a third-party language model to propose cleaning rules. The current production provider is DeepSeek. TidyX is an independent service and is not affiliated with, endorsed by, or sponsored by DeepSeek. Providers may change, but the data-minimization and human-review controls described on this page will continue to apply.

Human oversight

Because no AI output is ever executed without an explicit human confirmation step, and because the executing engine only accepts allowlisted operations, the model cannot execute changes on its own. Impact figures shown during plan review are preliminary AI estimates, not a deterministic preview or a guarantee of the number of records that will change. After execution, the run report records actual changes measured by the deterministic engine. You remain responsible for checking the proposed rules and validating the resulting data.

What you may not use this service for

You may not upload or process:

  • Data you have no legal right to process, including personal data you are not authorised to handle.
  • Content that is unlawful, or whose processing would breach applicable data protection, export control or sanctions law.
  • Malware, or files crafted to attack the service or other users.
  • Data intended to build profiles for unlawful discrimination, surveillance of individuals, or any use prohibited by applicable law.

You may not attempt to make the model produce content outside the cleaning-rule schema, bypass the confirmation step, or extract other users' data.

Reporting and enforcement

Suspected misuse can be reported to support@tidyx.io. We may suspend accounts that breach this policy.

Changes

Material changes to this policy will be posted on this page.