The Atlantic AI Watchdog: How to Check If Your Content Was Used to Train AI Models

If you are a musician, writer, filmmaker, or any kind of creator, there is a good chance your work has been used to train an AI model — without your knowledge, without your permission, and without any payment.

The Atlantic has built a free tool to help you find out.

AI Watchdog, launched in 2025 and expanded to include music in June 2026, lets creators search large datasets that AI companies have used — or referenced — in their model development. You type in your name or the title of your work, and the tool tells you whether it appears in any of the indexed datasets.

This guide explains exactly what AI Watchdog is, how to use it, what the results actually mean, and what your options are if your work shows up.

What Is The Atlantic AI Watchdog?

AI Watchdog is an investigative tool built by The Atlantic magazine, compiled by researcher Alex Reisner. It was originally launched in 2025 as part of a broader investigation into the content AI companies used to train their models — focusing first on books, research papers, and video content.

In June 2026, The Atlantic expanded the tool to cover music, adding a searchable database that draws on four large music datasets commonly referenced in AI research and development. These datasets collectively contain tens of millions of tracks — including recordings from major label artists, independent musicians, and underground producers across virtually every genre.

The tool is free to use and requires no account or signup.

What Can You Search?

AI Watchdog currently covers content in several categories.

Books and written works were the first category covered when the tool launched in 2025. If you are an author, you can search for your name or book title to see if your published works appear in datasets like Books3 or other large text corpora used in language model training.

Research papers and academic content are also indexed — relevant for academics and researchers concerned about their published work being used in AI training without attribution or compensation.

Video content was added in a later expansion, allowing filmmakers and video creators to search whether their work appears in video datasets used to train AI video generation models.

Music is the most recent addition, added June 2026. The music database draws on four datasets spanning millions of tracks — with the two largest containing approximately 12 million and 9 million recordings respectively, alongside two smaller archives with over 100,000 recordings each.

How to Use The Atlantic AI Watchdog — Step by Step

Using the tool is straightforward. Here is the exact process.

Step 1 — Go to The Atlantic AI Watchdog

Open your browser and go to The Atlantic’s AI Watchdog page. The direct URL is theatlantic.com/technology/archive/ai-watchdog. The tool is free and accessible without a subscription.

Step 2 — Select your content category

The tool has separate search sections for different content types — books, music, video, and research papers. Navigate to the section that matches what you want to search. For musicians, go to the music section. For writers, go to the books section.

Step 3 — Enter your search terms

For music, you can search by artist name, band name, or track title. For books, search by author name or title. The search is not case-sensitive and handles partial matches, so you do not need to type an exact title.

Step 4 — Review your results

The results will show whether your work appears in any of the indexed datasets and which specific dataset it appeared in. Each result includes the dataset name and, where available, additional details about the specific entry.

Step 5 — Note what the results mean

This is the most important step — understanding what a positive result actually tells you. A match means your work appears in a dataset that has been referenced in AI research or shared in AI development communities. It does not definitively confirm that your work was used to train any specific AI model.

What Do the Results Actually Mean?

This is where many creators get confused — and where it is important to be precise.

If your work appears in the tool:

Your recording or written work is present in at least one of the datasets that the tool has indexed. Companies like Google and Stability AI have openly acknowledged using some of these datasets. However, the AI industry is opaque about exactly which datasets each model was trained on — so a match does not prove that any specific company used your work.

If your work does not appear:

This does not mean your work was not used in AI training. The tool only covers four music datasets — there are many other datasets in circulation that are not yet indexed. An absence from the tool’s results is not evidence of absence from all AI training data.

What this means practically:

The tool is most useful as an investigative starting point rather than a definitive legal document. If your work appears, you have evidence that it was in a dataset with documented AI connections — which is meaningful context for any legal or licensing conversation you might want to have. It is not, by itself, proof of infringement.

Why Did The Atlantic Build This?

The AI Watchdog project grew out of a broader investigation The Atlantic conducted into AI training data practices. The investigation found that AI companies had used enormous amounts of copyrighted content — books, music, videos, research papers — in training their models, often without informing the creators whose work was included, obtaining licenses, or providing compensation.

This practice sits at the heart of one of the most contested legal and ethical questions in the AI industry right now. AI companies have generally argued that training on publicly available data constitutes fair use under copyright law. Creators and rights organizations have argued the opposite — that using their work as training material, at commercial scale, requires licenses and compensation.

The legal battles are ongoing. The Recording Industry Association of America (RIAA) has taken action against AI music platforms. Music rights organizations in Germany and other EU countries have won landmark rulings establishing that commercial AI training requires licensing agreements. The Atlantic’s AI Watchdog gives individual creators a way to see where they stand within this broader dispute — and potentially to contribute to the documentation of how widespread unlicensed AI training data collection actually is.

If you are interested in the broader legal picture, the best free AI tools guide on Sigma Wire covers many of the platforms whose training data practices are currently under scrutiny — useful context for understanding which tools may be built on contested data.

What Can You Do If Your Work Appears?

Finding your work in the AI Watchdog results is not the end of the road — it is potentially the beginning of a process. Here are the practical options available to creators.

Document everything

Take screenshots of your results, noting the dataset name and the specific entry. Record the date you conducted the search. This documentation could be relevant if you pursue any legal or licensing avenue later.

Check if a licensing avenue exists

Some AI companies have introduced licensing programs for creators — particularly in the music industry following pressure from rights organizations. Check whether the companies associated with the datasets where your work appeared have any licensing or compensation programs in place. This landscape is changing rapidly in 2026.

Connect with relevant rights organizations

If you are a musician, your performing rights organization — ASCAP, BMI, SESAC in the US; PRS in the UK; GEMA in Germany — may be tracking AI training data use and pursuing collective licensing negotiations on behalf of their members. Contact them with your findings.

Consider legal advice if you have significant commercial works

If you are a professional creator with commercially significant works appearing in AI training datasets, a legal consultation with an intellectual property attorney may be worthwhile. The legal landscape around AI training data is active — new rulings are coming in regularly from courts in Germany, the US, and the EU — and an attorney can advise you on what standing you might have based on current precedent.

Use the finding as context in licensing conversations

If you are a creator who licenses your work professionally, evidence that your work has already appeared in AI datasets gives you documented leverage in any conversation about AI licensing terms with companies that approach you.

The Bigger Picture — Why This Matters for All Creators

The AI Watchdog is not just a tool — it is part of a larger shift in how creators are understanding and responding to AI’s relationship with their work.

For decades, the economics of the internet worked in one direction: platforms took content, monetized attention, and the creators whose content drove that attention received little direct compensation. AI has accelerated that dynamic dramatically. The datasets that trained the most capable AI models in the world were built from human creative work — and most of the humans whose work contributed to those datasets were never asked, never told, and never paid.

AI image generation tools, music generation platforms, and writing assistants — including many of the most capable free AI tools available in 2026 — are built on this foundation. The legal and ethical questions about whether that foundation was built fairly are being worked out in courts, in legislatures, and in public debate right now.

Tools like The Atlantic AI Watchdog give individual creators a window into a process that was previously entirely opaque. Whether that translates into compensation, licensing, or legal remedy depends on developments that are still unfolding. What it does is make the invisible visible — which is the necessary first step.

Limitations of the Tool

Being clear about what AI Watchdog cannot do is as important as understanding what it can do.

It only covers four music datasets. The tool’s music database draws on four specific datasets that researcher Alex Reisner identified in research papers and AI data-sharing sites. There are many other datasets in circulation — some of them much larger — that are not yet indexed. The tool is not comprehensive.

It cannot confirm which AI models used your work. Even if your track appears in a dataset, the tool cannot tell you whether OpenAI, Google, Apple, Stability AI, or any other company actually ingested that dataset when training a specific model. Dataset availability and model training are not the same thing.

It does not provide legal evidence of infringement. The tool’s results are a research starting point, not a legal finding. An attorney would need to help you assess whether the presence of your work in a dataset constitutes actionable infringement under applicable law.

Coverage will expand but is always incomplete. The Atlantic has stated its intention to expand AI Watchdog’s coverage as new datasets are identified. But the AI training data landscape is large, partly hidden, and continuously evolving. Any snapshot of what is indexed will always lag behind reality.

Frequently Asked Questions