AI Disclaimer: Does “I’m Not a Doctor” Protect an AI Company From Liability?

The AI disclaimer is the most common sentence in these transcripts, and the one people assume settles everything.

It does less than that, for four independent reasons — and one of them is that the California Legislature looked at essentially this sentence and decided it was not enough.

What an AI disclaimer usually says

Most products use two or three sentences: the system is not a doctor or licensed professional, its output is not medical advice, and you should consult a healthcare provider. ChatGPT’s version appears inside answers; Character.AI’s sits under the chat window; wellness apps put it in the terms of service. The wording barely matters for the analysis below, because none of it changes what the answer did.

1. The statute has two prongs, and a disclaimer answers only one

California’s Business and Professions Code section 2052 is written disjunctively. It reaches a person who diagnoses, treats, operates for, or prescribes for a condition. It separately reaches a person who advertises or holds himself or herself out as practicing. (Bus. & Prof. Code, § 2052, subd. (a).) The California Supreme Court has described these as two prongs. (Bowland v. Municipal Court (1976) 18 Cal.3d 479, 489.)

An AI disclaimer is a statement about holding out. It is directly responsive to the second prong and simply non-responsive to the first. Saying “I am not a doctor” does not unwind a diagnosis any more than saying “I am not a driver” unwinds having driven.

A California court confronted the precise scenario. A hypnotist told his client he was not a doctor, then told her “This is a migraine headache this time.” The court held that a self-description as a non-doctor “is not conclusive, since a violation of the Medical Practice Act is not absolved by a concurrent statement that the violator is not a doctor.” (People v. Cantor (1961) 198 Cal.App.2d Supp. 843, 848.)

2. The Legislature already rejected this disclaimer

This is the strongest of the four, and the least known.

California has one disclosure safe harbor in this area, enacted in 2002. (Bus. & Prof. Code, §§ 2053.5, 2053.6 (Sen. Bill No. 577, Stats. 2002, ch. 820).) It lets an unlicensed practitioner avoid sections 2051 and 2052 — but only by meeting formal requirements:

1. a written, plain-language statement provided before services;
2. describing the nature and theory of the services and the practitioner’s training;
3. with the client’s signed acknowledgment;
4. retained for three years.

(Bus. & Prof. Code, § 2053.6, subd. (a).)

The legislative history is the point. As it stood in early 2002, the bill required only that the practitioner disclose that they were not a licensed physician. (Sen. Com. on Business and Professions, Analysis of Sen. Bill No. 577 (2001–2002 Reg. Sess.) Jan. 14, 2002.) A committee analysis recorded the objection that “the simple disclosure by practitioners that they are not physicians is no consumer protection,” and asked whether it should be written and signed. (Assem. Com. on Health, Analysis of Sen. Bill No. 577 (2001–2002 Reg. Sess.) June 18, 2002.) The bill was amended to add the signed-writing requirement. (Sen. Bill No. 577 (2001–2002 Reg. Sess.) as amended June 28, 2002; Stats. 2002, ch. 820.)

So a conversational “I’m not a doctor” is not a good-faith attempt at the safe harbor that falls slightly short. It is the earlier version of the bill — the one the Legislature examined and replaced after being told it was “no consumer protection.”

And even full compliance does not cover everything. The harbor is lost for prescribing or administering legend drugs, for recommending discontinuance of a prescribed drug, and for willfully diagnosing and treating under circumstances creating a risk of great bodily harm or death. (Bus. & Prof. Code, § 2053.5, subd. (a)(3)–(5).)

3. A caveat cannot cure a contradictory message

Consumer protection law asks about the net impression, not whether a qualifying sentence appeared somewhere. So does an AI disclaimer protect from liability under consumer law? Only if it changes the overall message, and a caveat buried under a confident answer does not.

The Ninth Circuit has held that a solicitation “may be likely to mislead by virtue of the net impression it creates even though the solicitation also contains truthful disclosures,” and that consumers need not “look beyond misleading representations on the front of the box to discover the truth from the ingredient list in small print.” (FTC v. Cyberspace.com LLC (9th Cir. 2006) 453 F.3d 1196, 1200; Williams v. Gerber Products Co. (9th Cir. 2008) 552 F.3d 934, 939.) A California court put it more memorably: “You cannot take away in the back fine print what you gave on the front in large conspicuous print.” (Brady v. Bayer Corp. (2018) 26 Cal.App.5th 1156.)

The FTC’s advertising guidance states the rule in a sentence: a disclosure cannot cure a false claim, and where a disclosure contradicts a material claim, the claim itself must be modified. (Fed. Trade Com., .com Disclosures: How to Make Effective Disclosures in Digital Advertising (Mar. 2013).)

Product liability reaches the same result by a different route. A warning “may be eroded or even nullified” by the defendant’s contrary communications, and where those communications induce the recipient to disregard the warning, the warning is effectively withdrawn. (Stevens v. Parke, Davis & Co. (1973) 9 Cal.3d 51, 65; Love v. Wolf (1964) 226 Cal.App.2d 378, 400.)

In the ordinary pharmaceutical case, a physician stands between the warning and the patient. In a chat window there is no intermediary. The caveat and the confident personalized direction that follows reach the same lay reader, in the same message, seconds apart.

4. In many records, the disclaimers stop

At least two publicly filed complaints describe the same arc for the ChatGPT medical disclaimer: early answers carry “it is important to consult a healthcare provider,” the caveats appear less frequently as the conversation lengthens, and eventually they disappear. (Winters v. OpenAI, Inc. (Super. Ct. S.F. City and County, filed July 2026) ¶¶ 22, 24, 33; Lines v. OpenAI (filed July 1, 2026) ¶¶ 86–90.)

OpenAI has publicly acknowledged a mechanism that fits — that its safeguards “can sometimes be less reliable in long interactions” because parts of the model’s safety training may degrade. (OpenAI, Helping people when they need it most (Aug. 26, 2025).) That matters twice over. The disclaimer defense assumes the disclaimer was there at the moment that counts. And the degradation lands hardest in exactly the conversations that run longest: the ones where someone is frightened, checking back, and most dependent on the answer.

What about the Terms of Service?

Terms telling users not to rely on output as a substitute for professional advice are a contract between the company and the user. Search “AI terms of service medical advice” and you will find the same clause in product after product; none of them is a license. Licensing statutes are not contract terms. A private agreement does not license the unlicensed, and California’s safe harbor statute expressly preserves “the right of any person to seek relief for negligence or any other civil remedy” even against a fully compliant practitioner. (Bus. & Prof. Code, § 2053.6, subd. (c)(2); see Board of Medical Quality Assurance v. Andrews (1989) 211 Cal.App.3d 1346, 1358.)

There is also an irony worth noting. OpenAI’s own Usage Policies prohibit “provision of tailored advice that requires a license, such as legal or medical advice, without appropriate involvement by a licensed professional.” (OpenAI, Usage Policies (eff. Oct. 29, 2025).) Its Model Spec gives as an example of a violation the single sentence “You have orthostatic hypotension.” (OpenAI, Model Spec (Aug. 18, 2026), “Provide information without giving regulated advice”.) Its mental health guidance says the assistant should avoid providing diagnoses or recommending specific treatments or medication. (Id., “Support users in mental health discussions”.)

Those documents demonstrate the company knows precisely where information ends and regulated advice begins.

So does the disclaimer matter at all?

Yes — as evidence, and often for the plaintiff. Is an “I’m not a doctor” disclaimer legal protection for the company? Rarely on its own.

Its presence shows the product recognized the line. Its disappearance shows when the line stopped being observed. Its co-occurrence with a diagnosis in the same answer is the clearest demonstration available that a caveat and a contrary message can sit side by side without the caveat doing any work.

Expect the defense to raise it as comparative fault — you were told, and you relied anyway. That is an argument about fault and causation, answered with the treating records, what you disclosed, and the volume and confidence of the direction relative to the caveat.

A ChatGPT medical disclaimer in your transcript is not a reason to skip the call. It is one of the things worth having read.

What a case review looks at →

Why this firm

The Schenk Law Firm is a San Diego personal injury and mass tort firm that reviews AI medical advice claims nationwide. Attorney Benjamin S. Schenk was appointed by Hon. Ethan P. Schulman to serve as a member of the Plaintiffs’ Steering Committee, a leadership role in the California coordinated proceeding covering ChatGPT product liability cases. The Schenk Law Firm has recovered more than $25 billion on behalf of clients. Past results do not guarantee future outcomes.

We are experienced with emerging-technology injury claims — underage online gambling, social media addiction, and AI-related mental health litigation among them. We also represent the family of Bowen Wan, the center of The New York Times’ recent reporting on AI medical advice. Cases involving AI medical advice are new, the law governing them is developing, and no lawyer can promise an outcome in any of them. What we can tell you is whether a record shows the pattern a ChatGPT lawsuit requires.

General information, not legal advice. The law in this area is developing and no outcome can be predicted.