Skip to content

AI companies

How does Meta source training data?

By SourceX Editorial · Updated

Short answer

Meta says AI at Meta, including Llama-based features, is trained on a mix of publicly available data, licensed data, and information from Meta's own products and services, including publicly shared Facebook and Instagram posts. Meta has also faced copyright litigation alleging use of pirated book datasets. This page reflects Meta's disclosures and reported litigation, not an independent review of Meta's training corpus.

Meta's stated data sources#

Meta's AI transparency center says the information used to develop and improve AI at Meta includes a mix of publicly available data, licensed data, and information from Meta's products and services, including publicly shared posts from Instagram and Facebook and people's interactions with AI features.

Public posts used for Meta AI#

Reuters reported in 2023 that Meta trained parts of its Meta AI assistant on public Facebook and Instagram posts, while excluding posts shared only privately with family and friends.

  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

Open model releases#

Meta has released Llama models, including Llama 3, as openly available models, alongside safety tools such as Llama Guard, without itemizing a detailed training data source list in its public blog announcements.

Copyright litigation#

Authors including Sarah Silverman and Richard Kadrey filed a class action complaint against Meta in 2023 alleging that Llama was trained using unauthorized copies of their copyrighted books, a case that remains part of the public record on Meta's training practices.

What this means for a business#

A business that has posted content publicly on Meta platforms, or is considering a licensing relationship, should understand that public posts can be treated differently than private content, and should review privacy settings and licensing terms before assuming any specific category of content is excluded from AI training.

SourceX and Meta#

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

Frequently asked questions

Does Meta train AI on public Facebook and Instagram posts?

Yes, Reuters reported in 2023 that Meta trains parts of its Meta AI assistant on public Facebook and Instagram posts while excluding private posts.

Has Meta been sued over training data sourcing?

Yes, a group of authors filed a class action against Meta in 2023 in the Northern District of California alleging unauthorized use of copyrighted books.

Does Meta license data from outside companies?

Meta's transparency center states that AI at Meta training data includes licensed data in addition to public and product data.

Sources

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

  1. Meta's stated data sources: public data, licensed data, and product/service data transparency.meta.com
  2. Meta AI trained on public Facebook and Instagram posts, excluding private posts, reported 2023 reuters.com
  3. Meta Llama 3 release as an openly available model ai.meta.com
  4. Authors' class action complaint alleging Llama trained on unauthorized copyrighted books storage.courtlistener.com

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