Business owner sues OpenAI over ChatGPT's allegedly faulty legal guidance

A new lawsuit filed in the Northern District of California claims OpenAI is liable for allegedly inaccurate business and legal advice generated by ChatGPT.

According to the complaint, which was filed by attorney Jeremy Kimmelman of The Law Offices of James L. Arrasmith, OpenAI falsely represented that ChatGPT could serve as a capable business assistant, provide reliable advice and draft usable documents.

The plaintiff, who owns a bridal design business, says she purchased a premium ChatGPT subscription in 2025 to help with business strategy and document drafting.

"Plaintiff reasonably expected, based on Defendant's presentation of ChatGPT as an assistant and Defendant's commercial sale of access to the service/product, that ChatGPT would provide materially accurate, non-deceptive, and appropriately qualified output, especially when asked legal, business, financial, contractual, and/or confidentiality-related questions," the complaint states.

The case is Craddock v. OpenAI OpCo LLC, 3:26-cv-06920 (N.D. Cal., filed July 7, 2026).

The plaintiff says she first ran into problems last summer when she asked ChatGPT to help her contact manufacturers. ChatGPT generated confidentiality agreements that were "incomplete, inconsistent, or insufficiently qualified," which she then sent to manufacturers, according to the complaint.

"Plaintiff alleges that ChatGPT's incomplete or inconsistent guidance caused her to believe she had legal and business protections that were materially more robust than they were, thereby creating commercial risk and delay," the complaint states.

OpenAI could not be reached for comment on the lawsuit.

In July 2025, the plaintiff says she had an interaction with ChatGPT that lasted from midnight until 5 a.m., during which she questioned the AI assistant's truthfulness and asked it to provide only accurate information going forward. She then asked the chatbot to sign contracts stating it would tell the truth and would not disclose confidential information. The chatbot responded with what it represented as an electronic signature that it claimed was binding under the Electronic Signatures in Global and National Commerce Act of 2000, the plaintiff alleges.

According to the complaint, the plaintiff reasonably relied on that representation because OpenAI "intentionally created the conditions under which ChatGPT communicated, and reasonably appeared to communicate, authoritatively in the first person ... thereby fostering the appearance of human-like authority and inducing reasonable consumer reliance."

The company "knew or should have known paying users would reasonably rely" on ChatGPT's output, including legal advice, the complaint alleges. As a result, the plaintiff claims her business suffered harm and her intellectual property was left vulnerable by the faulty confidentiality agreements.

The plaintiff claims OpenAI failed to warn customers that ChatGPT does not provide reliable advice and that the company made both negligent and intentional misrepresentations about the model's capabilities. She also alleges violations of California and Colorado consumer protection laws, in addition to a breach of contract claim, arguing the company should be bound by the contract the chatbot attempted to sign.

The complaint follows a series of other lawsuits addressing whether OpenAI is liable for ChatGPT's output, including allegations that the chatbot encouraged someone to commit suicide or ingest a dangerous substance that led to an overdose.

"This is part of a wave of lawsuits in which people try to hold AIs responsible when harms happen after a user interacts with an LLM," UC Law San Francisco professor Robin Feldman told the Daily Journal. "The topics vary considerably, but the core question remains the same. Are AI models more like libraries with information or more like active advisors or participants?"

According to Feldman, a court will balance the alleged defect identified by the plaintiff against the effect a potential fix would have on the product. Courts also must consider whether the plaintiff used the product properly, she said.

"Other key issues will include whether this is a defective product or just a product that was used improperly and whether the warnings are sufficient," Feldman said. "Many [large language models] contain a blanket warning up front saying, 'This is an AI. AIs make mistakes' or specific warnings saying, 'I am not a doctor and cannot give medical advice.' Did such warnings exist, and, if so, are they sufficient?"

The case is assigned to U.S. Magistrate Judge Alex G. Tse in San Francisco.