Superhuman used journalists' names without permission — its CEO explains himself on air

Grammarly, renamed Superhuman, launched a feature called Expert Review that generated editorial advice in the names of real journalists and experts, without their knowledge or consent. The "experts" included The Verge editor Nilay Patel and investigative journalist Julia Angwin. The scandal grew big enough that Angwin filed a class action lawsuit, and Superhuman CEO Shishir Mehrotra agreed to be interviewed by Patel, a person whose name his company had used without permission.

Superhuman is an American AI company that renamed its legal entity from Grammarly in late 2025. Grammarly's flagship product is familiar to roughly 40 million daily users as an AI writing assistant. The company also develops Coda (documents), the Mail email client, and the Superhuman Go platform, a set of personalized AI agents that work directly inside the apps an employee already uses.

The Expert Review feature appeared in August 2025. It offered users editorial suggestions "inspired by" specific media figures, naming them and linking to their publications. Nobody was asked for permission.

In the interview, Mehrotra admitted the feature was a failure, both for users and for the experts. It was barely used, did not fit the company's strategy, and was switched off before the lawsuit was filed. Superhuman's first reaction, however, was to offer an email opt-out rather than remove the feature immediately.

The sharpest moment of the interview was the discussion of attribution and impersonation. Mehrotra insisted that Expert Review did not engage in impersonation: the feature was merely "inspired by" specific people's work and said so clearly. Patel pushed back: what the system generated in his name had nothing to do with his actual editorial advice. "I never talk about emotional headlines. The Vergecast is not a show about editing smartwatch headlines." In his view, attributing someone's text is not the same as putting someone's name on generated content that the person would never have produced.

Angwin's lawsuit is based not on impersonation but on New York and California laws that prohibit using people's names and identities for commercial purposes without consent. Mehrotra maintains that the claims have no merit, and notes that no LLM today obtains permission for attribution. He does concede, however, that the legal minimum is not the standard to aim for. His benchmark is the YouTube model, which he helped build: a platform creators will want to join of their own accord, with a clear revenue share. Superhuman Go plans to share 70% of revenue with the authors of agents.

The conversation went beyond a single incident. Patel cited an NBC News poll in which public perception of AI is worse than that of the immigration service. In his view, the reason is the extractive nature of the technology: AI takes other people's work as raw material and offers nothing of equal value in return. Mehrotra countered that people fear job losses, not attribution problems. Superhuman's goal, he said, is not to replace people but to expand their capabilities and turn them into "superhumans."

This episode is a symptom of a systemic contradiction. AI companies build products on human creativity without establishing a clear economic deal with creators. The courts have not yet answered the key question of whether training LLMs on other people's texts infringes copyright. While that question remains open, scandals like this are inevitable.

Superhuman shut down the feature and apologized, but the lesson for the whole industry is much broader.