When a team buys an AI model or coding tool, the security questionnaire usually asks about data. The answers differ by product tier, not just by vendor. A chat plan, an API account and an enterprise contract can sit under three different sets of terms.
This post lists questions to put to any AI vendor, and shows what three vendors publish on their own pages as of 18 October 2026, so you know what a normal answer looks like. It is not legal advice, and your contract overrides any public page.
Start with the product tier
Ask which exact product and plan you are buying, and get the data terms for that tier in writing. Public pages show how much terms vary:
- Anthropic's pricing page lists model training as "Opt-out" for Free, Pro and Max, "No model training on your content by default" for Team, and "None by default" for Enterprise.
- Anthropic's commercial terms say "Anthropic may not train models on Customer Content from Services," and that the customer "retains all rights to its Inputs, and (b) owns its Outputs."
- OpenAI's API documentation says "data sent to the OpenAI API is not used to train or improve OpenAI models" unless you explicitly opt in.
- Google's Gemini API terms treat paid and unpaid services differently: paid services do not use prompts to improve products, while unpaid services do and may involve human review.
The questions
Data use
- Is our content used to train or improve any model, by default, under this plan? Is that in the contract, or only on a web page that can change?
- Is there any human review of our prompts or outputs, and under what conditions?
- Do you use our data for abuse monitoring, and is that data used for anything else?
- Do outputs belong to us, and who is liable if an output infringes?
Retention
- How long are inputs and outputs stored, by default, and where?
- How long are chats and coding sessions kept in the product, and how is deletion handled?
- Do you offer zero data retention, what does it exclude, and who approves it?
- What is kept longer when content is flagged by safety systems?
For baselines: Anthropic's privacy center says standard API inputs and outputs are automatically deleted from backend systems "within 30 days of receipt or generation," that deleted chats leave back end storage within 30 days, and that different terms such as a zero data retention agreement can apply. It also says flagged inputs and outputs can be kept for up to 2 years and classification scores for up to 7 years. OpenAI says abuse monitoring logs are retained for up to 30 days by default, unless longer retention is required by law, and that approved customers may select Modified Abuse Monitoring or Zero Data Retention, "subject to prior approval by OpenAI." Some endpoints are not ZDR eligible and "may still store application state." Google's paid terms say prompts and responses are logged "for a limited period of time" for detecting violations, without stating the period on the page I read, and that data may be stored transiently or cached in any country where Google or its agents have facilities.
Location and access
- In which regions is data processed and stored? Can we pin a region, and is there a price for it?
- Who at the vendor and its subprocessors can access our content, and how is that logged?
- Which subprocessors are used, and how do we get notice of changes?
On residency pricing, Anthropic's page lists a 1.1x multiplier for US only inference and a 10 percent premium for regional and multi region endpoints on some cloud platforms. Treat such details as price terms to negotiate, not afterthoughts.
Controls and audit
- What admin controls exist: SSO, SCIM, role based access, audit logs, retention settings? OpenAI lists SAML SSO for Business and SCIM, EKM, RBAC and a Compliance API for Enterprise and Edu.
- Can we export our conversation and usage records?
- What independent assurance reports can you share under NDA, and for which product?
Money and exit
- What is billed per seat and what is billed per token? Anthropic lists Enterprise at $20 a seat with a 20 seat minimum, "plus usage billed at API rates."
- What happens to our data at contract end, and in what time?
- Can we switch models or vendors without rewriting our integration?
How to use the answers
Put the answers in your vendor register, not an email thread. Record the plan name, the date, the document you relied on, and who accepted the risk. Re-check on renewal, because public terms and prices move.
Question 17 is where architecture helps. If your tooling lets you bring your own model, a data term you cannot accept becomes a configuration change instead of a migration. Safeguard supports bringing your own model, including on premises and local options, and its MCP tool calls are unlimited.
Quick check
Is a vendor's public web page enough to rely on for data terms?
No. Public pages describe defaults and can change. Get the terms for your exact plan in the signed contract or data processing addendum.
What does zero data retention usually require?
An agreement with the vendor. OpenAI says controls are subject to prior approval, and Anthropic says the standard 30 day window does not apply where a zero data retention agreement exists. Endpoints can be excluded, so ask which ones.
