There are two different stories behind claims that Apple and Anthropic used YouTube creators’ work for AI: a 2024 report about caption text in a training corpus, and a 2026 lawsuit alleging Apple used a video dataset to retrieve creators’ videos. Neither establishes that both companies downloaded complete YouTube videos for AI training or that a court has found either company liable.
What the 2024 report said about YouTube captions
WIRED and Proof News reported in 2024 that the YouTube Subtitles dataset, containing caption text associated with YouTube videos, was included in The Pile, a large text corpus. The report counted 173,536 videos from more than 48,000 channels. That is the outlet’s reported count, not a settled audit figure.
As an Amazon Associate I earn from qualifying purchases.
The distinction matters: the reporting concerned subtitle and caption text, not proof that Apple and Anthropic each downloaded and trained on complete audiovisual files directly from YouTube. Anthropic confirmed that The Pile had been used for Claude. Its spokesperson, Jennifer Martinez, said, “The Pile includes a very small subset of YouTube subtitles,” as quoted by WIRED. Apple did not respond to that investigation’s request for comment; WIRED reported that Apple research materials described use of The Pile.
There is a count discrepancy in later legal material. A 2025 amended complaint in Concord Music Group v. Anthropic states that the dataset contained 173,651 videos, rather than the 173,536 reported by WIRED and Proof News. The complaint’s number is a claim in a court filing, not an independent resolution of the discrepancy. WIRED and Proof News’ 2024 report and the amended complaint describe the figures.
#1 Best Overall
Creators’ reaction is not a court finding
Creators interviewed for the report described discovering their work in the dataset without being asked. David Pakman, host of The David Pakman Show, said, “No one came to me and said, ‘We would like to use this,’” Julie Walsh Smith, CEO of Complexly, said the company was “frustrated to learn” its educational content had been used without consent, and Professor Dave Explains host Dave Farina argued that compensation or regulation should be discussed if work is used to build products that could displace creators. These are creators’ reported experiences and views; they do not by themselves determine whether a particular use violated copyright or another law.
How the 2026 Apple lawsuit is different
A separate proposed class action filed in April 2026 names Apple and concerns Panda-70M, a video dataset—not the YouTube Subtitles component of The Pile. The plaintiffs, including Ted Entertainment and the owners of MrShortGame Golf and Golfholics, allege that their videos were accessed through Panda-70M and that their content appeared more than 500 times in that dataset. The complaint alleges video retrieval and circumvention of YouTube protections; those claims remain allegations in the sources reviewed.
Rank #2
In a July 2026 response, Apple argued that the videos were publicly available and that accessing them was permitted by the DMCA and YouTube’s Terms of Service. That is Apple’s litigation position, not a court ruling. The reports reviewed do not establish a decision on the merits or a final disposition. See the MacRumors coverage of Apple’s response and the case docket for the filings and status information.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Two datasets, two different claims
| Question | 2024 YouTube Subtitles / The Pile reporting | 2026 Apple / Panda-70M lawsuit |
|---|---|---|
| What material is described? | Subtitle and caption text associated with YouTube videos, included in a text corpus. | A video dataset said in the complaint to identify or contain videos and clips. |
| Which company connection is documented? | Anthropic confirmed The Pile was used for Claude; reporting linked Apple research materials to The Pile, while Apple did not comment to the investigation. | Apple is the defendant named in the reported proposed class action. |
| What is the evidence status? | Investigative reporting, research references and Anthropic’s comment; the reported video count differs from a later complaint’s count. | Plaintiffs’ allegations and Apple’s response; no merits ruling is established by the cited coverage. |
| What is the central issue? | Whether creators’ caption text was included in training material without their awareness or permission. | Whether Apple unlawfully accessed source videos or bypassed YouTube protections. |
What Apple says about its training data and crawler
In a September 9, 2026 disclosure, Apple described training data as coming from several sources, including publicly available web-crawled information, directly licensed or purchased data, open-source data, user-study data and synthetic data. Apple says Applebot does not crawl sites that require login credentials or are protected by paywalls. It also says the crawler respects standard robots.txt directives publishers can use to ask it not to crawl their sites or not to use website content for foundation-model training, and describes filtering and processing steps.
Rank #3
This is Apple’s general account of its practices. It does not establish how a particular third-party dataset was obtained, whether a specific video or caption appeared in one, or whether a use was lawful. Apple’s disclosure is available on its Applebot and training-data page.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What is established—and what is not
- Established in the cited reporting: The YouTube Subtitles dataset was included in The Pile, and Anthropic confirmed The Pile was used for Claude.
- Reported, but not proof of full-video downloading: Apple research materials described use of The Pile. The 2024 report focused on caption text.
- Alleged in a separate lawsuit: Apple accessed creators’ videos through Panda-70M and circumvented YouTube protections. Apple disputes the legal premise, arguing public availability and lawful access.
- Not established by these sources: a final court determination that either company infringed copyright, violated YouTube’s terms, or unlawfully obtained the material.
Apple’s August 2025 STIV paper describes a text- and image-conditioned video-generation method, but it does not show that The Pile’s YouTube Subtitles material trained STIV. The STIV paper is a separate piece of context, not evidence connecting that caption dataset to the system.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
