Alex Reisner

Alex Reisner is a staff writer at The Atlantic who investigates how generative AI is built on vast stores of unlicensed creative and journalistic work, using technical reporting and original data tools to show what and who is inside these systems. His coverage centers on AI training data and the hidden pipelines that move books, music, video, and paywalled news into commercial models without consent. He works at the intersection of technology, publishing, and creative labor, combining programming and reporting to make opaque AI practices legible to the people whose work is being used.

Tracing AI’s Use of Books and Hollywood Writing

Reisner’s defining work documents how large language models are trained on massive collections of pirated books and screenwriting, naming specific datasets and the systems that rely on them.

In an early investigation into generative AI, he revealed that more than 190,000 pirated books had been used to train prominent models, turning what had been an abstract debate over “training data” into a concrete list of affected authors. He has followed that thread through subsequent reporting on the scale of AI’s book problem, showing how internal deliberations at major companies moved from considering licenses to explicitly choosing mass infringement as a faster path.

His coverage extends to Hollywood, detailing how dialogue from movies and television has been harvested for AI, again pairing narrative reporting with searchable datasets that allow writers to see whether their work appears in training corpora.

Reporting on AI’s Impact on Music, Video, and Publishing

Beyond books and scripts, Reisner tracks how AI companies are targeting other creative industries, including recorded music and online video. In “The Millions of Songs Mashed Into AI-Generated Music,” he shows how developers rely on vast collections of pirated tracks to produce AI-generated songs that closely mimic the originals, and he publishes tools and audio comparisons that expose specific overlaps between training data and outputs.

His reporting on AI systems built from millions of YouTube videos catalogues the companies involved and frames the resulting products as direct competitors to the creators whose work supplied the training material, again accompanied by a dataset search tool that lets people check whether their videos were used. He also writes about the broader consequences for publishers, arguing in a recent piece on the end of publishing as it is currently structured that AI is subjecting books and journalism to the same kind of platform shock that ride-hailing imposed on taxi businesses.

Examining Technical Failures and Memorization in Generative AI

Reisner’s beat is not limited to copyright and consent; he also dissects how generative AI actually works at a technical level and what that means for users and regulators. In “AI’s Memorization Crisis,” he reports on research showing that these systems do not learn in a human sense but instead store and reproduce training data, and he presents evidence that companies have tried to downplay or hide this behavior. His article “How 4chan Gamers Accidentally Invented AI ‘Reasoning’” traces current claims about chatbot reasoning back to a specific online community’s techniques, connecting marketing language to a concrete, and much narrower, underlying trick. In “Generative AI Is an Engineering Disaster,” he shifts to infrastructure and efficiency, arguing that the industry has built systems that are fundamentally wasteful and that the resulting demand for memory and energy imposes real costs on people who are not direct users of AI products.

AI Watchdog Project and Data-Driven Method

Reisner founded The Atlantic’s AI Watchdog project, a continuing effort to investigate the data and practices behind AI systems rather than their consumer-facing features. The initiative has produced articles paired with interactive tools that allow readers to search training datasets for their own names or works, closing the gap between high-level policy discussions and personal stakes. His reporting on Common Crawl, for example, shows how a little-known nonprofit has scraped paywalled journalism from major outlets and supplied it to AI companies, and it links that finding to specific archives and technical workarounds used to evade publisher restrictions.

Across this project, he writes in a restrained, technical style, foregrounding documentation, code, and primary data so that affected authors, artists, and publishers can verify his findings for themselves.