Skip to content

AI companies

How does Amazon source training data?

By SourceX Editorial · Updated

Short answer

Amazon Web Services says its generative AI services are trained on licensed and proprietary datasets, synthetic datasets, open-source datasets, and publicly available content including web-crawled data. Amazon has also signed content licensing deals, including with the New York Times for use in products like Alexa. This page reflects Amazon's own disclosures and reported deals.

AWS's generative AI disclosure#

AWS's Generative AI Development Disclosure states that AWS trains and tests its generative AI services on licensed and proprietary datasets, synthetic datasets, open-source datasets, and publicly available content, including web-crawled data, which may include rights-protected material and in some cases personal information incidentally included in the data.

Amazon foundation model privacy page#

Amazon's Bedrock privacy documentation says Amazon's foundation models use licensed, open-source, proprietary and publicly available data, and that AWS customer prompts and fine-tuning data are not used to train the underlying Amazon foundation models.

  1. 01Your work records
  2. 02Personal details removed
  3. 03You approve the release
  4. 04Licensed to AI developers
How it works with SourceX: Your work records → Personal details removed → You approve the release → Licensed to AI developers.Illustrative

The New York Times deal#

Reuters reported in May 2025 that the New York Times signed a multi-year deal letting Amazon use its editorial content, including NYT Cooking and The Athletic, for AI products such as Alexa and for training Amazon's proprietary foundation models, without disclosing financial terms.

Transparency index findings#

Stanford's Foundation Model Transparency Index reports on Amazon's Nova models note that Amazon discloses using licensing, web crawling, synthetic generation and human annotation as data acquisition methods, without naming specific top sources by volume.

What this means for a business#

A business evaluating any AWS-hosted AI service should check whether it is using Amazon's own foundation models or a third-party model hosted on Bedrock, since each has distinct data handling commitments, and should confirm contractually that uploaded business data is excluded from model training.

SourceX and Amazon#

SourceX is not affiliated with Amazon. This page summarizes public reports; it does not mean Amazon buys data through SourceX.

Frequently asked questions

Does Amazon use customer prompts to train its AI models?

Amazon's Bedrock privacy page states that customer prompts and outputs are not used to train the underlying Amazon foundation models, with limited exceptions it discloses.

Has Amazon signed content licensing deals for AI?

Yes, Reuters reported in 2025 that the New York Times licensed content to Amazon for use in products including Alexa and for training Amazon's foundation models.

Does Amazon disclose exactly what web data it crawls?

No. AWS's disclosure describes categories such as licensed, proprietary, open-source and publicly available web-crawled data, but does not name specific sources.

Sources

Public sources, checked 2026-10-09. SourceX is not affiliated with the organizations named.

  1. AWS generative AI training data categories: licensed, proprietary, synthetic, open-source, publicly available web-crawled data aws.amazon.com
  2. Amazon foundation models not trained on customer prompts/fine-tuning data aws.amazon.com
  3. New York Times licensing deal with Amazon for Alexa and foundation model training, reported 2025 reuters.com
  4. Amazon Nova data acquisition methods per Stanford transparency index crfm.stanford.edu

Related resources

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies Talk to SourceX
See if your company qualifies