What If AI Gives the Wrong Drug or Dosage Advice and Someone Is Injured?

If an AI system gives someone the wrong drug, wrong dose, wrong timing, or dangerous interaction advice and the person is physically injured after following it, there may be a viable Utah injury claim. The legal theory will depend on what the AI was designed to do, who built or deployed it, whether a licensed health professional was involved, and how the recommendation reached the patient. A case may involve product liability, ordinary negligence, medical malpractice, consumer-protection law, or several of those theories at the same time. Utah has not yet produced a definitive appellate ruling that every generative-AI chatbot is, or is not, a “product,” so the technology and the facts matter more than the marketing label.

A consumer chatbot that invents a dangerously incorrect dosage is a different liability problem from a hospital system that recommends a dose inside an electronic prescribing workflow, but both can cause the same medical harm. A medication error can lead to overdose, undertreatment, toxic drug interactions, bleeding, organ injury, cardiac complications, hospitalization, permanent impairment, or death. In either setting, the claim turns on whether the bad recommendation can be preserved and traced to the injury rather than treated as an isolated screenshot with no context. That makes early evidence preservation unusually important in AI medication cases.

Wrong AI Medication Advice Can Create More Than One Liability Path

The first question is not simply whether the AI was wrong; it is who was responsible for putting the dangerous recommendation into the decision chain. A general-purpose AI company may have designed a system that gives highly specific medication instructions despite known reliability limits, while a health system may have embedded a clinical tool into patient care and encouraged clinicians to rely on it. A physician, pharmacist, hospital, software vendor, platform operator, or other entity may each have a different role, insurance policy, defense, and set of records. In a serious case, identifying those roles early can determine which evidence must be preserved before each defendant begins pointing at the others.

Consider a patient who asks an AI assistant whether to increase a medication and receives a precise dosage that is unsafe for the patient’s age, kidney function, or other prescriptions. Now consider a hospital decision-support system that reads a laboratory value incorrectly, converts units incorrectly, or recommends a dose without accounting for an interaction already documented in the chart. The first scenario may focus heavily on the design and warnings of a consumer AI product or service, while the second may also raise questions about professional standards of care and the institution’s deployment of the technology. The same injury can therefore produce very different claims depending on how the recommendation was generated, presented, reviewed, and acted upon.

Utah Product Liability Law May Apply, but the AI “Product” Question Is Not Settled

Utah’s Product Liability Act provides that a product cannot be considered defective unless, when it was sold by the manufacturer or other initial seller, it contained a defect or defective condition that made it unreasonably dangerous to the user or consumer. Utah defines “unreasonably dangerous” by reference to what an ordinary and prudent buyer, consumer, or user would expect, while also considering the particular user’s actual knowledge, training, or experience. Utah law recognizes design defects, manufacturing defects, and inadequate warnings as traditional product-liability theories. For an AI medication case, design-defect and failure-to-warn theories may be the most natural fit because the alleged problem can involve the system’s decision architecture, safety guardrails, data sources, validation, or presentation of medical advice rather than a one-off manufacturing flaw. Utah Legislature

That does not mean Utah law has already decided that every AI system is a “product.” The current Utah Product Liability Act does not supply a general definition of “product,” and no Utah appellate decision appears to have resolved whether a modern generative-AI chatbot falls inside the Act simply because it is software. In Blaisdell v. Dentrix Dental Systems, Inc., the Utah Supreme Court dealt with purchased dental software and a strict-products claim, but the decision ultimately enforced a contractual limitation of liability rather than announcing a broad rule that all software is a product. A Utah AI case would therefore need to address the actual software, distribution model, subscription or sale relationship, updates, intended use, and alleged defect rather than assume the classification question is settled. Utah Legislature

Courts elsewhere are beginning to confront that issue, but the law is still developing. In Garcia v. Character Technologies, Inc., a federal court in Florida held at the motion-to-dismiss stage that the Character.AI app could be treated as a product to the extent the claims attacked alleged design defects rather than ideas or expressions inside the app. The case settled in January 2026, so it did not produce a trial verdict or appellate merits decision, and it is not binding on a Utah court. It nevertheless shows why an AI injury case should separate a claim about dangerous product design from a complaint that the plaintiff simply disliked information or speech generated by software. FindLaw

Utah Law Already Treats Some AI Medical Advice as High Risk

Utah has also enacted AI-specific consumer rules that are directly relevant to medical advice, even though those statutes do not automatically decide a personal-injury lawsuit. Utah Code Title 13, Chapter 77 defines a “high-risk artificial intelligence interaction” to include personalized recommendations, advice, or information that could reasonably be relied on for significant personal decisions, including medical advice or services. An individual providing services in a regulated occupation must prominently disclose certain high-risk generative-AI interactions and must still comply with the requirements of the regulated occupation when providing services through AI. Utah law also states that it is not a defense to specified consumer-protection violations that generative AI made the statement, undertook the act, or was used to further the violation. Utah Legislature

Those provisions matter because Utah has already rejected the idea that adding AI to a transaction necessarily erases ordinary legal responsibility. At the same time, Chapter 77 is primarily a consumer-protection and disclosure regime, and its existence should not be confused with a blanket private cause of action for every harmful AI output. The chapter expressly says that it does not displace other rights or remedies under state or federal law, which leaves traditional tort, malpractice, contract, and product-liability questions to be analyzed on their own terms. In a medication-injury case, the AI statute may therefore provide useful context about risk and disclosure without replacing the need to prove duty, defect or breach, causation, and damages. Utah Legislature

When a Doctor, Pharmacist, or Health System Uses AI

When a doctor, pharmacist, hospital, clinic, or other Utah health care provider uses AI as part of patient care, the case may also fall under Utah medical-malpractice law. The central question becomes whether the human provider and institution met the applicable professional standard of care when they selected, interpreted, checked, or acted on the AI recommendation. A clinician cannot necessarily avoid scrutiny by saying “the computer told me to do it,” just as an AI vendor is not automatically relieved of responsibility merely because a clinician was somewhere in the chain. Expert review is often needed to determine what a reasonably careful provider should have verified before prescribing, dispensing, changing, or administering the medication.

Medication cases often involve overlapping failures rather than a single bad actor. The AI may have generated an incorrect dose, the electronic health record may have transmitted the wrong unit, the prescriber may have failed to notice an implausible recommendation, and the pharmacy may have missed a warning that should have triggered review. Defendants and their insurers may each argue that someone else broke the causal chain, which is why the complete workflow matters more than the final prescription alone. A careful investigation should reconstruct who saw what information, at what time, on which screen, under which software version, and what each person did next.

FDA Regulation Can Matter Without Deciding the Utah Injury Claim

Federal regulation can also matter, but FDA status does not by itself answer who is civilly liable in Utah. The FDA’s January 2026 Clinical Decision Support Software guidance explains that some decision-support software may fall outside the federal “device” definition when statutory criteria are met, while other software functions—including some intended for patients or caregivers—can remain subject to medical-device regulation. FDA guidance also emphasizes the importance of whether a health care professional can independently review the basis for a recommendation rather than relying primarily on the software. Those distinctions can become important evidence about intended use, risk controls, validation, labeling, and what the developer expected users to do. U.S. Food and Drug Administration

A general consumer chatbot that happens to answer a medication question is not automatically an FDA-regulated medical device, and a hospital dosing tool is not automatically exempt simply because it uses AI. The regulatory analysis is function-specific, which means investigators need to know exactly what the software was intended to do and how it was marketed and deployed at the time of the injury. Regulatory compliance can be relevant to a defect or negligence analysis, but it should not be mistaken for a complete defense to every state-law claim. Likewise, the absence of FDA regulation does not by itself prove that a developer or provider acted reasonably. U.S. Food and Drug Administration

Proving That the AI Error Actually Caused the Injury

To build a strong case, the plaintiff must connect the digital error to the medical injury with evidence rather than inference alone. That usually means proving the exact recommendation, showing that the patient or provider actually relied on it, establishing what medication was taken or administered, and obtaining medical evidence that the incorrect drug or dose caused the claimed harm. In an overdose case, records may include emergency treatment, toxicology, laboratory trends, antidote administration, cardiac monitoring, dialysis, intensive care, or other treatment consistent with the alleged exposure. In an undertreatment case, the medical question may be whether the AI-induced delay or dose reduction allowed the underlying condition to worsen in a way that probably would not otherwise have occurred.

Causation is where defendants will often try to turn a straightforward error into a complicated medical dispute. They may point to the patient’s underlying disease, other prescriptions, alcohol or supplements, prior symptoms, inconsistent dosing, or a later medical decision and argue that one of those factors—not the AI recommendation—caused the outcome. That makes the timeline critical, because the claim should show the sequence from prompt or clinical input to recommendation, reliance, medication exposure, symptoms, treatment, and diagnosis. Qualified medical and technical experts may both be necessary when the mechanism of injury and the software behavior are outside ordinary knowledge.

Preserve the AI Evidence Before It Changes or Disappears

The AI conversation itself should be preserved as evidence, not merely summarized from memory. Keep the full conversation thread, screenshots, account-export data, timestamps, model or product identifiers, app or browser version, email confirmations, notifications, and any screen recording that shows how the recommendation appeared in context. If the AI system cited a source, opened a health-information card, or displayed a warning before or after the harmful recommendation, that context should be saved too because the defense may later argue that the user ignored qualifying information. Do not assume the same prompt will reproduce the same answer later, because generative systems, retrieval sources, safety rules, and model versions can change.

The physical and medical evidence matters just as much as the digital evidence. Preserve the prescription bottle, label, packaging, written instructions, remaining medication when appropriate, pharmacy receipt, prescriber order, discharge paperwork, and records showing when the medication was obtained or administered. In a provider-based case, the electronic health record, medication-administration record, pharmacy system, audit trail, order history, clinical messages, and decision-support alerts may reveal who saw or changed the recommendation. Counsel can also send targeted preservation demands seeking relevant vendor logs, model-version records, system prompts or safety configurations, validation materials, incident reports, and other records before routine data retention or software updates make reconstruction harder.

How Technology Companies and Insurers May Defend an AI Medication Case

Technology companies, health systems, and insurers have several predictable ways to narrow an AI medication claim. They may argue that the user should never have relied on a chatbot, that a disclaimer defeated any reasonable reliance, that a physician or pharmacist became the sole cause, that the software was a service rather than a product, or that the injury came from the underlying condition instead of the bad recommendation. An online platform may also raise federal defenses such as Section 230 or constitutional arguments depending on how the claim is framed, although Section 230 by its text addresses information provided by another information content provider and does not resolve every claim involving a platform’s own alleged product design or generated output. The response is not to debate those defenses in the abstract; it is to preserve the architecture, warnings, output, workflow, medical evidence, and causal chain well enough that the case can be decided on what actually happened. Legal Information Institute

Insurance carriers may also try to obtain a recorded statement, broad medical authorization, early release, or quick settlement before the technical and medical evidence has been developed. That is particularly risky when there may be multiple defendants, future complications, disputed software responsibility, health-insurance liens, or a release broad enough to extinguish claims against parties not yet identified. A serious medication injury should not be valued solely from the initial hospital bill or from the first defendant’s view of who is at fault. Liability, prognosis, future care, insurance coverage, liens, and release language should be understood before the claim is resolved.

Damages and Filing Deadlines Can Make Early Investigation Important

Damages in a successful AI medication case may include emergency and hospital expenses, future medical care, lost income, reduced earning capacity, disability, pain, loss of enjoyment of life, and other harms recognized by Utah law. If the medication error causes death, the case may also involve Utah wrongful-death and estate claims, with a different damages analysis and additional procedural questions. The medical consequences can continue long after the acute overdose or interaction has resolved, especially when there is organ damage, neurological injury, or permanent functional loss. The damages presentation should therefore be built from the full prognosis rather than the earliest snapshot of treatment.

Time limits can also differ depending on the theory and the defendant. Utah’s Product Liability Act generally requires a product-liability action to be brought within two years after the claimant discovered, or with due diligence should have discovered, both the harm and its cause, while Utah medical-malpractice law generally uses a two-year discovery period with a four-year outer limit subject to statutory exceptions. A mixed AI case can involve additional statutes, procedural prerequisites, contractual issues, or defendants whose deadlines do not match each other. Prompt investigation is important not only because of filing deadlines but because software logs, account data, pharmacy records, and institutional audit trails may be easier to preserve now than months later. Utah Legislature

Talk to a Utah Attorney About an AI Medication Injury

An AI medication injury is not just a “bad answer” case; it is a medical-causation case, a technology-evidence case, and potentially a product-liability or malpractice case at the same time. The strongest investigation starts by preserving the exact AI interaction, identifying every human and corporate actor in the medication chain, and obtaining the records needed to test each defendant’s explanation. Gabriel K. White and The Legal Beagle represent injured people and families in Utah serious-injury, product-liability, medical-malpractice, and wrongful-death matters. Call The Legal Beagle at (801) 915-6152 or contact the firm at https://www.mylegalbeagle.com/contact.