AI Training Collides With Identity and Copyright Law

AI companies are building models on top of people’s work and even their identities, then arguing that this is just how innovation works. A growing backlash is testing whether the law—and the market—will force those firms to seek consent and share the upside, or cement a world where creative labor is scraped, cloned and remixed for free.

The fight is no longer abstract. It is arriving as specific lawsuits, leaked lobbying memos and new bills over names, voices and faces.

A lawsuit over “AI experts” who never signed up

On March 11, investigative journalist Julia Angwin filed a federal class-action lawsuit in New York against Superhuman Platform, the company behind Grammarly, accusing it of misappropriating her name and the identities of other writers through its “Expert Review” feature. The suit alleges that Grammarly presented AI‑generated feedback as if it came from real journalists and academics, including Angwin, without informing them or securing consent, to sell a premium service to users. Wired and Angwin’s lawyers describe it as “identity theft” in service of AI.

The same day the suit was filed, Superhuman disabled Expert Review and said it would not bring the tool back in its current form, acknowledging that “some of our communications were not as clear as they should have been” about how the feature worked, according to reporting from Wired and the complaint itself. A company spokesperson has said Grammarly never trained its core models directly on the private documents of users but has been more circumspect about how it drew on publicly available writings from named “experts.” The case now turns on the right of publicity—the legal protection for a person’s name and likeness—rather than traditional copyright alone.

Angwin’s lawsuit lands amid a broader wave of litigation by authors, visual artists and news outlets arguing that generative AI systems were trained on their work without a license. Those suits, including high‑profile cases against OpenAI and Meta, contend that mass scraping of books, articles and images turns copyright into a one‑time “data grab” rather than an ongoing right to control and profit from creative work, as outlets like The New York Times and Reuters have reported.

Lobbying to redefine copyright for the AI era

Even as courts begin to weigh those claims, parts of the industry are pushing to rewrite the rules. In Australia, a leaked policy paper from the Business Council of Australia proposed amending the country’s Copyright Act so that using creative works for AI training would no longer count as infringement—effectively creating a broad “free use” zone for training data. Investigative outlet Crikey reported that the idea was quietly dropped after a backlash from artists and after the federal government publicly ruled out such a change.

Similar arguments have surfaced in the United States and Europe, where major AI labs and tech trade groups have urged regulators to treat model training on copyrighted material as “fair use” or as a permitted text‑and‑data mining exception, warning that stricter limits would undercut domestic AI capabilities. OpenAI, for example, has argued that training on copyrighted works is both legal and essential to remain competitive with China, telling policymakers that its model training “aligns with the core objectives of copyright and the fair use doctrine,” according to reporting from Ars Technica.

Lawmakers are responding unevenly. In the U.S., proposals like the Generative AI Copyright Disclosure Act would force companies to list the copyrighted works used in training but stop short of banning such use, while state‑level laws such as Tennessee’s ELVIS Act target unauthorized AI impersonations of a person’s voice or image. Analyses in the Journal of Intellectual Property Law & Practice argue that transparency alone will not resolve the underlying tension between creators and AI developers over what, if anything, must be licensed or paid for.

Who owns an identity in the age of training data?

What makes the Grammarly case uniquely volatile is that it blends these copyright fights with a direct assault on the idea that anyone controls their own professional persona. Expert Review didn’t just train on articles; it presented an AI’s output as if it were an “Angwin‑style” review, raising the specter of AI systems that can be hired out as stand‑ins for living professionals while the humans remain uncompensated and unaware. Coverage in eWEEK notes that similar consent battles are already playing out over deepfake porn and AI‑cloned celebrities.

For creatives, journalists and ordinary users, the stakes are now clear. One path normalizes the idea that anything posted or recorded can be ingested into models and repackaged under someone else’s brand, with at most after‑the‑fact transparency. Another path insists on consent, licensing and enforceable rights of publicity, even if that slows down or fragments AI development. Courts and legislators are still deciding which vision wins.

In the meantime, the Angwin suit and the leaked Australian lobbying push function as warnings. They show how quickly AI firms will test the edges of identity and copyright—and how urgently societies will need new rules, and perhaps new economic institutions, to ensure that the people whose words and likenesses train the machines are more than raw material.