Writing Tool Removes AI Personas After Legal Challenge from Authors

Grammarly has disabled an AI feature that mimicked the writing styles of prominent authors and scientists without their consent, following a significant legal challenge from the writers whose identities were used. The Expert Review function, which offered writing feedback “inspired by” the personas of figures including Stephen King and Carl Sagan, was taken down this week by Superhuman, the tech firm that operates Grammarly. The move came after a multi-million dollar lawsuit was filed in the Southern District of New York by investigative journalist Julia Angwin and other writers who discovered their names and professional reputations being marketed as commercial AI personas. Superhuman’s chief executive acknowledged the tool had “misrepresented” the voices of experts|expert voices, apologizing for the controversial feature.

The Feature That Sparked Public Backlash

The Expert Review capability was a substantial departure from Grammarly’s conventional editing support approach. Rather than providing one-size-fits-all feedback, the tool enabled users to obtain editing suggestions “inspired by” the distinctive approaches of renowned writers and academics. Users could choose from personas including highly-regarded novelist Stephen King and prominent scientist Carl Sagan, among hundreds of other public figures. The feature offered to provide tailored writing advice examined through the lens of these distinguished experts, ostensibly assisting users enhance their work by drawing inspiration from the best in their respective fields.

What Grammarly promoted as an innovative educational tool soon became apparent as a troubling misappropriation of identity and intellectual property. The company had not sought permission from any of the writers whose personas were being copied and sold. Investigative journalist Julia Angwin, who led the case in the collective legal action, voiced concern at discovering her professional reputation being sold as a product feature. She described the situation as substantially distinct from traditional deepfakes, emphasizing that her editing abilities is her means of earning and that she had never imagined her professional skills could be stolen and packaged this way.

• AI personas emulated hundreds of writers without authorization or financial compensation
• Feature promised suggestions based on renowned writers and researchers
• Users could pick different expert personas for editing suggestions
• Tool was built into Grammarly’s paid subscription service offerings

Legal Action and Market Reaction

The legal challenge against Superhuman and Grammarly marks a pivotal point in the wider conversation over ethics in AI and copyright protections. Led by journalist investigator Julia Angwin, the class-action lawsuit filed in the Southern District of New York alleges that the company illegally appropriated the identities of numerous writers to create revenue from its subscription-based service. The filing argues that utilizing identities and professional reputations for business use without explicit consent infringes upon established law protecting individuals from unauthorized exploitation of their identity and image.

The response to the lawsuit has been swift and substantial. Within 24 hours of submission, Angwin’s legal team indicated receiving from over 40 potential plaintiffs keen to participate in the action, illustrating broad apprehension among affected writers. The case pursues damages surpassing $5 million, though industry analysts indicate the actual figure could be considerably higher once the court determines compensation derived from the company’s earnings from the disputed feature. Superhuman’s quick move to deactivate the Expert Review function suggests the company recognized the legal and reputational risks posed by maintaining the feature.

The Court Case Details

The lawsuit maintains that Grammarly and Superhuman infringed upon essential safeguards of identity protection by crediting writing guidance to authors that did not furnish such advice. The court documents underscores that the company commercialized these personas through its subscription-based revenue system, earning income directly from the unauthorized use of hundreds of individuals’ names and professional standing. Attorneys argue this amounts to a “brazen violation of the law,” pointing to existing legal precedents protecting individuals from commercial misappropriation of their identities without approval.

Julia Angwin’s personal frustration with the feature extended beyond the unlawful infringements to the caliber of what the AI produced. She labeled the revision recommendations assigned to her as a “slopperganger”—a term denoting inferior AI-produced material—observing that the edits were making sentences inferior rather than enhancing them. This additional layer to the case emphasizes not only the regulatory violation but also the damage to professional standing of having one’s identity connected to below-standard professional content, intensifying the harm of improper use of one’s name.

• Damages sought surpass $5 million with final amount determined by company earnings
• More than 40 other claimants contacted law firm within the initial day
• Claims unauthorized commercial exploitation of identities lacking consent or compensation

Credibility Issues and Credibility Issues

Beyond the regulatory breaches, the Expert Review function raised serious questions about the reliability and accuracy of AI-generated editorial advice. Users depending on suggestions credited to recognized authors and scholars had no way of knowing whether they were receiving genuine guidance or algorithmically-generated approximations of professional expertise. This loss of confidence extends beyond single users to the broader writing community, where audiences and learners might have legitimately believed they were getting guidance from established authorities. The removal of the feature highlights a critical gap between what artificial intelligence can technically achieve and what it should ethically be permitted to do.

The reputational harm experienced by writers whose identities were used was especially problematic because it linked their names to substandard work. Angwin’s case illustrated the issue—her professional standing was being promoted as a high-quality offering while simultaneously delivering inferior editorial guidance. This pairing of unauthorized application and inferior standards resulted in twofold damage: lack of control over her name paired with association with inferior work that conflicted with her professional standards. For writers whose standing relies on the quality of their work, such false attribution poses an existential threat to their professional standing and market position.

The Problem with AI Imitation

The core issue in Grammarly’s methodology lay in seeking to mimic the nuanced judgment and skill of seasoned professionals through automated systems. Professional editing requires situational awareness, stylistic sensitivity, and extensive professional background—elements that cannot be authentically captured by examining text examples and producing edits in a similar voice. Angwin’s point that the AI-generated edits made sentences unnecessarily intricate rather than improving them revealed the essential inadequacy of the imitation. The platform could mimic obvious stylistic features but was missing the substantive grasp essential for delivering truly useful editing advice, ultimately weakening both the authenticity of the personas and the practical value of the offering itself.

Company Response and Path Forward

Superhuman’s chief executive Shishir Mehrotra acknowledged the misstep publicly, issuing an apology on LinkedIn in which he admitted that the Expert Review function had “misrepresented” the voices of the impersonated experts. The company’s quick move to turn off the feature this week suggests an effort to reduce additional legal and reputational damage. However, the deactivation came only after the lawsuit was filed and significant public backlash surfaced, raising questions about whether the company would have taken action without external pressure. Mehrotra’s statement, while apologetic in tone, did not tackle the larger issue of how such a feature was greenlit and launched in the first place, nor did it outline concrete measures to prevent similar incidents in the future.

The path forward for Grammarly remains uncertain as the lawsuit moves forward. Beyond the immediate legal challenge, the company faces the difficulty of regaining trust with both writers and users who may now challenge the ethical guidelines shaping its AI development. The elimination of the Expert Review function represents a reactive versus proactive stance, indicating the company is addressing legal pressure rather than exhibiting authentic dedication to ethical artificial intelligence. Moving forward, Grammarly will almost certainly need to introduce more stringent approval procedures and governance structures for any capabilities that employ the use of actual people’s identities. The company’s handling of this crisis may establish a precedent for how rival artificial intelligence companies approach the use of public figures in their AI tools.

The rapidity at which Grammarly deactivated the feature suggests the company understood the legal, reputational, and business stakes involved. However, the lack of proactive measures before the lawsuit demonstrates that internal review systems overlooked the violations of ethical standards. As the case moves forward, the company might experience further examination about how numerous content creators were impacted and whether damages will be provided above what the lawsuit demands. The case is likely to influence how other AI companies handle the incorporation of real identities in their products going forward.