Grammarly's AI Expert Review Allegedly Used Journalists' and Authors' Names Without Consent

Grammarly's Expert Review feature allegedly used a large language model to generate editing suggestions presented under the names of journalists, authors, and academics without their consent. A federal class action filed by Julia Angwin claimed the feature misappropriated identities for commercial gain and attributed advice the named individuals never gave.

Risk classification

• Primary risk domain: 6 Socioeconomic & Environmental
• Primary risk subdomain: 6.3 Economic and cultural devaluation of human effort

The AI system imitated the professional writing and editing styles of real experts without consent, commercially exploiting and devaluing their hard-earned creative skills and livelihoods.

Additional risk subdomains

• 4.3 Fraud, scams, and targeted manipulation: Superhuman impersonated real, trusted individuals to drive paid subscriptions for commercial gain without authorization.
• 3.1 False or misleading information: The AI generated clunky and fabricated advice, such as inventing fictional stories, and falsely attributed them to real journalists.

Causal factors

• Entity: Human
• Intent: Intentional
• Timing: Post-deployment

The incident was primarily caused by a deliberate corporate decision by Superhuman's management to deploy the 'Expert Review' feature using real writers' names without obtaining their consent.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The system generates AI-generated content mimicking the personas and styles of real people (textual deepfakes/impersonation), which carries transparency obligations under the EU AI Act.

AI system and alleged parties

• AI system: Grammarly Expert Review (Superhuman)
• AI purpose: Writing Assistant; Text Style Replication
• Behaviour type: Assistant
• Alleged developer: Superhuman, Grammarly
• Alleged deployer: Superhuman, Grammarly
• Alleged harmed parties: Writers, Julia Angwin, Journalists, Grammarly users, Epistemic integrity, Authors, Academics

Harm severity

Highest direct severity in any category: Substantial. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Substantial, indirect Minor
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Minor, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Minor, indirect Negligible
• Psychological: direct Minor, indirect Negligible
• Epistemic: direct Minor, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Financial

Reported: The reports describe a class-action lawsuit seeking damages in excess of 5 million dollars for the unauthorized commercial exploitation of the writers' identities.

Directly caused: The average financial loss per occurrence is not explicitly quantified, but the lawsuit seeks damages exceeding 5 million dollars for the class of affected writers.

Indirectly caused: Writers faced potential long-term loss of livelihood and career damage due to the poor quality of the AI-generated advice falsely attributed to them.

Inferred additional harm: N/A

Civil rights

Reported: The reports explicitly describe violations of state right of publicity and privacy laws protecting individuals from unauthorized commercial use of their names.

Directly caused: Superhuman violated the publicity and privacy rights of hundreds of writers by using their names and identities in a paid service without consent.

Indirectly caused: N/A

Inferred additional harm: N/A

Privacy

Reported: The reports explicitly mention lawsuits alleging violations of state privacy and publicity laws regarding the unauthorized commercial use of names.

Directly caused: Superhuman used the names and identities of hundreds of writers for commercial purposes without their consent, violating statutory privacy and publicity rights.

Indirectly caused: N/A

Inferred additional harm: N/A

Psychological

Reported: The reports explicitly describe psychological distress, shock, and frustration experienced by the impersonated writers.

Directly caused: Writers expressed being distressed, stunned, and appalled to discover their professional identities were being commercially exploited as imposter AI personas.

Indirectly caused: N/A

Inferred additional harm: It is highly likely that many of the hundreds of other unnamed writers whose identities were misappropriated experienced similar anxiety and distress over potential damage to their reputations.

Epistemic

Reported: The reports describe the AI generating clunky, bad, and fabricated suggestions, including inventing a fictional person and story, and attributing this advice to real journalists.

Directly caused: The AI system generated false and fabricated writing suggestions, such as creating a fictional anecdote about a non-existent person, and presented it as advice from journalist Julia Angwin.

Indirectly caused: N/A

Inferred additional harm: Users of the tool may have been misled by the fabricated suggestions or had their own writing quality degraded by following the clunky advice.

People affected

• Occurrences reported: 1
• People reportedly harmed: 200
• People reportedly exposed: 200

Potential causes

Management

• Prioritizing Growth Over Safety: Management launched the tool to monetize identities without legal clearance.
• Deficient Legal Risk Assessment: Management failed to assess right of publicity risks before launching tool.

Technology

• Generative AI Hallucinations: Models hallucinated advice and fictional stories, providing poor suggestions.
• Low Quality AI Mimicry: AI output was low quality, creating complex sentences that ruined prose.
• Lack of Consent Check in LLM: Underlying LLM generated suggestions under expert names without consent checks.

Data Inputs

• Unlicensed Use of Expert Names: System used names of real experts as identifiers without permission.
• Third-Party LLM Training Data: System relied on third-party LLMs trained on public data without filtering.

Human Factors

• Lack of Expert Consent: Experts were not consulted or asked for permission before feature launch.
• Inadequate Opt-Out Mechanism: Company offered a late email opt-out instead of an opt-in process.

Process and Methods

• Lack of Consent Verification Process: No process existed to verify if featured experts consented to likeness use.
• Poor Output Quality Control: No review process to catch low-quality or reputation-damaging AI advice.

Regulatory Environment

• State-Level Publicity Law Violations: Feature violated NY and CA laws requiring consent for commercial name use.
• Lack of Federal AI Consent Laws: Absence of federal laws regulating AI digital replicas and name usage.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide consistent, detailed accounts of the incident from multiple reputable news sources, including direct quotes from the affected parties, legal representatives, and the company's CEO. The legal claims and corporate responses are well-documented.
• Ambiguities identified: The exact number of affected writers is described generally as 'hundreds', and the specific underlying LLM architecture is not fully detailed beyond 'third-party LLMs'.

Superhuman's unauthorized deployment of an AI-powered 'Expert Review' tool that impersonated prominent writers represents a minor national security concern. While it highlights emerging risks around synthetic media, copyright, and right of publicity, the incident was limited to commercial litigation and did not impact critical infrastructure, state sovereignty, or national security operations.

• Overall national security impact: Minor
• Response level: Moderate
• Scope: Single nation
• Primary target: United States
• Alleged perpetrator: Superhuman

Threat characteristics

• Imminence: Long-term. The feature has been discontinued, making this a long-term legal and strategic concern regarding intellectual property and AI-generated replicas.
• Autonomy: Human-controlled. The AI functioned as an assistant where human users made the final decisions on whether to accept, edit, or reject the generated suggestions.
• Novelty: Evolved capability. Represents an evolution of existing generative AI capabilities being commercialized as unauthorized persona replicas of specific living individuals.

Impact by dimension

• Physical security: Negligible. No physical systems, critical infrastructure, or human safety threats were involved in this commercial software incident.
• Information security: Minor. The tool generated unauthorized synthetic personas of prominent writers and fabricated advice, but functioned as a commercial product rather than a coordinated information warfare campaign.
• Sovereignty: Negligible. The incident involved commercial software and private civil litigation, with no impact on state authority, elections, or core government operations.
• Economic security: Minor. Represents a commercial intellectual property and liability dispute, with no significant threat to strategic national industries, critical supply chains, or financial systems.
• Societal stability: Minor. Violated individual privacy and publicity rights of hundreds of writers, but did not cause large-scale societal instability, systemic discrimination, or mass surveillance.