The model is not a security boundary
The vendor acknowledges problems with precise calculation, dates, complex phrasing and irrelevant context. It also warns that malicious instructions in the input data can influence the model. Separate questions do not necessarily satisfy expected logical identities. S07
The safeguards we recommend: handle calculations and permissions in code, stop when a result is unclear, and keep the proposal separate from its execution. A message the model has classified is not automatically a resolved complaint.
A contract, not just a slogan
The Master Customer Agreement distinguishes between customer data and telemetry. TypeSafe will not include customer data in training its weights without prior consent. However, it may process telemetry (technical logs, metrics, classifications and aggregate statistics) without restriction, including to improve its services. Purchased credits expire at the end of the contract term or 12 months after purchase, whichever comes first. S13
The privacy policy mentions the hosting of the service and the transfer of data to the USA. For personal data, you also need to read the Data Processing Addendum and the terms of any intermediary platform. S14 S15
The legal index page mentions a zero data retention option for enterprise customers. Without verification, it cannot be assumed to be the default setting for every account. S12
Designing a safe pilot
Start with public product data or thoroughly anonymized examples. Do not upload an entire customer profile when the task only needs a short message. Decide who sees the result, how long it is kept and who can trigger an action.
This text is a technical and informational overview, not a legal confirmation of GDPR compliance. Before any real processing of personal data, the specific data flow must be assessed.
Sources and level of verification
- S07Jev 1.13 jaggedness ↗
TypeSafe AI · Primary source retrieved · check attempted 23 September 2026
- S13Master customer agreement ↗
TypeSafe AI · Primary source retrieved · check attempted 23 September 2026
- S14Privacy policy ↗
TypeSafe AI · Primary source retrieved · check attempted 23 September 2026
- S15Data processing addendum ↗
TypeSafe AI · Primary source retrieved · check attempted 23 September 2026
- S12Legal ↗
TypeSafe AI · Primary source retrieved · check attempted 23 September 2026