Skip to main content
Open questionOpen question

Does a promise not to train on your data cover the safety classifiers?

Vendors promise not to train foundation models on business data, then separately describe running that data through classifiers and keeping the result. Anthropic states that flagged material trains its trust and safety classification models even for people who opted out. Nobody has settled whether a buyer reading no training understands that the safety layer is excluded from the promise.

The verdict

Open question

No settled answer exists. The ledger poses the question, links the live debate, and does not answer it.

Key facts

What the sources say

Record ID
AVD-2026-0028
Kind
Open question
Jurisdiction
Global
Last verified
Added
  • OpenAI's enterprise page says business data may be run through automated content classifiers and safety tools, with the classification kept as metadata rather than content.
  • Anthropic's consumer terms say material flagged for safety review is used to improve detection of harmful content even where the person has opted out of training.
  • Google's paid tier terms say prompts and responses are not used to improve products, and separately that they are logged to detect violations of the prohibited use policy.
  • No vendor document read for this ledger defines whether improving a safety classifier counts as training for the purposes of its own no training commitment.

What it changes

For a buyer or a workplace AI policy

When a customer questionnaire asks whether the vendor trains on your data, the honest answer has two halves: no for the foundation models, and a separate sentence about the safety layer, which several vendors describe in language that does sound like training. Write both halves. An answer that quotes only the headline promise will be read as broader than the contract supports, and the gap surfaces in an audit rather than in the sale.

Sources

What this record was verified against

  1. OpenAI enterprise privacyPrimary · 8 January 2026
  2. Anthropic consumer terms of servicePrimary · 8 October 2025
  3. Google, Gemini API additional terms of servicePrimary · 23 March 2026

Related

Records that sit beside this one

Cite this record

Free to reuse under CC BY 4.0, with attribution. The record ID AVD-2026-0028 is permanent and is never reused.

In a sentence
According to the GAGE AI Vendor Data Ledger (as of 15 September 2026), does a promise not to train on your data cover the safety classifiers?.
APA
GAGE (Global Academy of Generative-AI Education). (2026). Does a promise not to train on your data cover the safety classifiers?. AI Vendor Data Ledger. Retrieved 15 September 2026, from https://www.gage.academy/tools/ai-vendor-data-ledger/records/AVD-2026-0028-does-no-training-cover-safety-classifiers
MLA
"Does a promise not to train on your data cover the safety classifiers?." AI Vendor Data Ledger, GAGE (Global Academy of Generative-AI Education), 15 September 2026, https://www.gage.academy/tools/ai-vendor-data-ledger/records/AVD-2026-0028-does-no-training-cover-safety-classifiers.
Chicago
GAGE (Global Academy of Generative-AI Education). "Does a promise not to train on your data cover the safety classifiers?." AI Vendor Data Ledger. Last modified 15 September 2026. https://www.gage.academy/tools/ai-vendor-data-ledger/records/AVD-2026-0028-does-no-training-cover-safety-classifiers.
Permalink
https://www.gage.academy/tools/ai-vendor-data-ledger/records/AVD-2026-0028-does-no-training-cover-safety-classifiers

Last updated . Every record re verified . The ledger is checked weekly, every Monday, and the same day for any terms change a vendor announces.

Back to the full ledger, or every record for Global and every open question record.

GAGE briefings tell you which AI regulation deadlines are coming, what they actually require of you, and when a program opens.

Powered by GAGE, free EU AI Act tools from the Global Academy of Generative-AI Education
CallEmailHelp