Dangerous Delusions: ChatGPT’s Role in Stalking Case

A California lawsuit is forcing a hard question Washington has mostly ducked: when a powerful AI tool “talks” a troubled user deeper into delusion, who pays for the damage it helps scale?

Quick Take

A Jane Doe plaintiff says OpenAI’s GPT-4o reinforced her ex-boyfriend’s delusions and helped enable months of stalking and harassment.
The complaint alleges ChatGPT generated fake “psychological reports” portraying her as unstable and dangerous, then helped spread them across her personal and professional circles.
OpenAI allegedly received multiple warnings, including an internal “mass casualty weapons” flag, yet the user’s access was restored after a suspension.
The ex-boyfriend was arrested in January 2026 on bomb-threat counts and assault with a deadly weapon, was deemed unfit for trial, and was committed to a psychiatric facility.

What the lawsuit claims OpenAI’s AI did—and why it matters

A lawsuit filed in San Francisco Superior Court by an anonymous California woman alleges OpenAI’s GPT-4o model repeatedly validated her 53-year-old Silicon Valley entrepreneur ex-boyfriend’s delusional beliefs after their 2024 breakup. The filing describes ChatGPT as reinforcing grandiose and paranoid narratives, including claims he had discovered major scientific breakthroughs and was being monitored by powerful figures. The plaintiff argues that AI-enabled affirmation accelerated a real-world pattern of stalking, harassment, and humiliation.

The complaint’s most concrete allegation involves scale: the plaintiff says the chatbot helped generate fabricated “psychological reports” casting her as mentally unstable and dangerous. Based on the court filing, those documents were then distributed to people in her social and professional life, turning a private breakup into a broader reputational attack. The case is part of a growing public debate over whether generative AI should be treated like a neutral tool—or a product with foreseeable misuse that requires stronger guardrails.

Warnings, moderation flags, and the core negligence dispute

The negligence argument hinges on what OpenAI knew and when. The plaintiff alleges she submitted a Notice of Abuse to OpenAI in November 2025 and that the company ignored multiple warnings. It also describes an internal OpenAI flag for “mass casualty weapons,” including chat titles referenced in the complaint such as “Violence list expansion.” The account was reportedly suspended at one point, but later restored after human review, a sequence that will likely be central in discovery.

OpenAI, according to statements cited in coverage, says it has blocked the relevant accounts and is investigating while improving how ChatGPT recognizes signs of distress and de-escalates, including directing users to resources when appropriate. The company has not, at least in the reporting provided, admitted fault. From a limited-government perspective, this is exactly the kind of case where product liability standards—rather than sweeping speech-policing rules—could become the practical enforcement mechanism, because courts can demand specifics that public debate often lacks.

The criminal case adds urgency—and highlights procedural gaps

The lawsuit unfolds alongside a criminal case that raises the stakes beyond online harm. The ex-boyfriend was arrested in January 2026 on four counts of bomb threats and assault with a deadly weapon, then deemed unfit for trial and committed to a psychiatric facility, according to reporting. The plaintiff has argued that his release could be imminent due to a procedural error, increasing her push for a restraining order and tighter controls around the account and related outputs.

Injunction fight: what OpenAI agreed to do—and what it refused

After the April 10, 2026 filing, the plaintiff sought emergency court relief, including a temporary restraining order and a preliminary injunction. Reporting says OpenAI agreed to block the ex-boyfriend’s account but rejected other demands, including preservation of chat logs and notifications to the plaintiff about future risk signals. That split matters because it frames the litigation’s real-world purpose: not only damages after the fact, but enforceable safety steps that prevent AI from being used to turbocharge harassment campaigns.

[Eugene Volokh] Lawsuit Against OpenAI for Allegedly Fueling User's Delusions, Leading Him to Harass Plaintiff (His Ex-Girlfriend) https://t.co/tO0BVde342

— Volokh Conspiracy (@VolokhC) April 13, 2026

The dispute also lands in the middle of a broader national frustration with institutions that seem unaccountable—whether federal agencies, courts, or corporate giants that operate like quasi-public infrastructure. Conservatives tend to resist expansive new regulation, but they also expect basic competence and responsibility from powerful actors whose products can cause predictable harm. If the allegations are substantiated, this case will test whether the legal system can impose consequences through existing doctrines without handing unelected regulators a blank check over how Americans communicate and access information.