Does OpenAI API train on your data?
OpenAI API by OpenAIModel APIs and platformsVendor's pages read on
Not by default, according to the vendor's published terms.
The vendor states that inputs are not used for training unless the customer opts in on API Platform. Read on 12 Nov 2026. The plan your company actually holds, and any negotiated terms, decide what applies to you.
A summary of what the vendor publishes, not legal advice, not an assessment of the vendor and not an endorsement. How this page was made.
What the vendor states, plan by plan
| Plan | Trains on your data | Default retention | Human review of inputs | Single sign-on |
|---|---|---|---|---|
| API PlatformAPI | Only if the customer opts in | Up to 30 days (abuse monitoring logs) | Abuse monitoring only | Unknown |
“Unknown” means the vendor's page that would answer this was not read, or could not be opened. It does not mean “no”.
API Platform
On API Platform, the vendor states that inputs are not used for training unless the customer opts in. Retention: up to 30 days (abuse monitoring logs). A zero-retention option is stated. The vendor states that people review inputs only to investigate abuse. Stated as available: data residency options.
- Trains on customer data
- Only if the customer opts in
“data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us)”
The vendor states this on platform.openai.com, read on 12 Nov 2026
- Default retention
- Up to 30 days (abuse monitoring logs)
Some endpoints and features store application state for longer; see the source.
“By default, abuse monitoring logs are generated for all API feature usage and retained for up to 30 days”
The vendor states this on platform.openai.com, read on 12 Nov 2026
- Zero-retention option
- Yes
Subject to prior approval by OpenAI.
“You can also request zero data retention (ZDR) for eligible endpoints if you have a qualifying use-case.”
The vendor states this on openai.com, read on 12 Nov 2026
- Human review of inputs
- Abuse monitoring only
Authorized employees may also access stored data for engineering support, abuse investigation and legal compliance.
“specialized third-party contractors who are bound by confidentiality and security obligations, solely to review for abuse and misuse”
The vendor states this on openai.com, read on 12 Nov 2026
- Data residency options
- Available to eligible customers (project-level setting)
The source lists the regions and which support regional processing.
“Data residency controls are a project configuration option that allow you to configure the location of infrastructure OpenAI uses to provide services.”
The vendor states this on platform.openai.com, read on 12 Nov 2026
Unknown for this plan: Deletion on request; Single sign-on (SSO); SCIM provisioning; Audit logs; Admin console; Customer-managed keys.
Attestations and company facts the vendor states
- SOC 2 Type II
- Yes
“Our API Platform has been audited and certified for SOC 2 Type 2 compliance.”
The vendor states this on openai.com, read on 12 Nov 2026
- ISO/IEC 27001
- Yes
Stated scope: the OpenAI API, ChatGPT Enterprise and ChatGPT Edu.
“OpenAI maintains ISO/IEC 27001:2022 and ISO/IEC 27701:2019 certifications”
The vendor states this on openai.com, read on 12 Nov 2026
- ISO/IEC 42001
- Yes
“OpenAI maintains an ISO/IEC 42001:2023 AI Management System covering OpenAI's consumer and business AI products and models”
The vendor states this on openai.com, read on 12 Nov 2026
- HIPAA BAA available
- Yes
Stated for the API Platform; ChatGPT for Healthcare is described separately.
“We are able to sign Business Associate Agreements (BAA) in support of customers' compliance with the Health Insurance Portability and Accountability Act (HIPAA).”
The vendor states this on openai.com, read on 12 Nov 2026
- GDPR DPA available
- Yes
“we are able to execute a Data Processing Addendum (DPA) with customers for their use of ChatGPT Business, ChatGPT Enterprise, and the API”
The vendor states this on openai.com, read on 12 Nov 2026
- Data processing location
- Servers in various jurisdictions (see source)
“on servers located in various jurisdictions”
The vendor states this on openai.com, read on 12 Nov 2026
- Model providers used
- OpenAI's own models
“The OpenAI API Platform gives developers access to powerful models like our GPT-5 models.”
The vendor states this on openai.com, read on 12 Nov 2026
“Yes” means the vendor's page says so, with the page linked. GhostAudit has not audited the vendor. Scope by product is on the vendor's page.
What remains unknown
- Deletion on request
- Single sign-on (SSO)
- SCIM provisioning
- Audit logs
- Admin console
- Customer-managed keys
- FedRAMP
- Company headquarters
None of these was guessed, inferred from another vendor or filled from memory. Ask the vendor, using the list below.
- Enterprise controls not verified. SSO, SCIM and audit logs have not been checked for this tool yet.
How to check OpenAI API in your own company
- 01Find out which plan each account is on. People who signed up with a personal address are on the personal terms, whatever the company bought.
- 02Open the account or administrator settings and look for the data-controls or model-training setting the vendor's page describes. Note what it is set to today and who is able to change it.
- 03Read the vendor's own page, linked under each fact below, and check that the sentence quoted here is still there. Pages change; the date shows when this one was read.
- 04Compare with your contract or order form. Where a data processing agreement or negotiated terms exist, they take precedence over the public pages.
- 05Record the plan, the setting and the date in your AI inventory, so that the answer exists the next time a customer or auditor asks.
What to ask the vendor
Neutral questions for the points the vendor's public pages left open. Ask for the answer in writing.
- 01How do we request deletion of our data, and within how many days is it completed, including backups?
- 02Is single sign-on available on our plan, and can sign-in with personal accounts be blocked for our domain?
- 03Is SCIM provisioning available, so that leavers lose access automatically?
- 04Are audit logs available to administrators, and can they be exported to our security tooling?
- 05Is there an administrator console where data controls can be set for every account in our organisation?
- 06Can we hold the encryption keys for our data?
- 07Do the statements on your public pages apply to our contract as written, or do negotiated terms override them?
Questions about OpenAI API
- Does OpenAI API train on your data?
- Not by default, according to the vendor's published terms. The vendor states that inputs are not used for training unless the customer opts in on API Platform. Read on 12 Nov 2026. The plan your company actually holds, and any negotiated terms, decide what applies to you.
- Is OpenAI API safe to use at work?
- GhostAudit does not rate tools. Whether OpenAI API is acceptable for your company depends on the plan you hold, the data people put into it and your own policy. This page records what the vendor states, with the wording and the source, so that decision can be made on facts: API Platform: only if the customer opts in (training on customer data).
- How long does OpenAI API keep data?
- The vendor states: Up to 30 days (abuse monitoring logs). This is the default the vendor's page describes for one plan; other plans may differ, and the table on this page shows each.
- Which attestations does the vendor of OpenAI API state?
- The vendor's pages state: SOC 2 Type II, ISO/IEC 27001, ISO/IEC 42001, HIPAA BAA available and GDPR DPA available. This is the vendor's statement, with the page linked; GhostAudit has not audited the vendor, and the scope by product is on the vendor's page.
- When were these facts checked, and how do I report an error?
- The vendor's pages were read on 12 Nov 2026. Every quotation was checked against the stored text of its source page on that date. To report an inaccuracy, email [email protected] with the page that shows the current wording.
How this page was made, and how to correct it
Every fact above is a statement found on a page OpenAI itself publishes. The page was fetched and reduced to text, the statement was recorded with its exact wording, and the quotation was checked against the stored text of the page. Quotations are at most 25 words. Search results, blogs and third-party summaries were not used.
These are summaries of statements the vendor publishes, as read on the date shown. They are not legal advice, not an assessment of the vendor and not an endorsement. Vendors change their terms: open the linked source and check your own contract before you rely on a statement.
Found an error, or are you the vendor? Report an inaccuracy with the page that shows the current wording. The page is re-read; the fact, its quotation and its date are corrected, or the fact is set to unknown if the page cannot be opened. The full method.
The vendor's official pages
- Privacy policyopenai.com
- Termsopenai.com
- Data processing agreementopenai.com
- Subprocessorsopenai.com
- Trust centretrust.openai.com · not readable automatically
OpenAI API is a product of OpenAI. GhostAudit is not affiliated with, endorsed by or sponsored by OpenAI.
Is OpenAI API in use at your company, and on which plan?
GhostAudit shows which AI tools your people use, from Google Workspace or Microsoft 365 and from the browser, and sets each one beside the terms on this page. It records tool names and time, never prompts.