How Gen AI is causing bodily injury

Can Generative AI cause bodily injury?

On the face of it, it seems absurd to think that words or videos from ChatGPT or another Gen AI tool can lead to bodily injury. For the purposes of this analysis, ‘bodily injury’ encompasses physical harm, sickness, or disease sustained by a person (including death), as well as emotional distress and mental anguish. This would include things like self-harm, suicide, poisoning, overdose, and certain psychological issues.

However, if we exclude robotics and other physical manifestations of Gen AI like vehicles or connected machines, Generative AI (via voice, text, images, etc.) can cause bodily injury arguably through both direct and indirect routes.

The causal logic is as follows:

Direct route: Gen AI output is advice or instruction → user trusts output and implements it → bodily injury of user.

Indirect route: Gen AI output is advice or instruction → reviewed by someone and forwarded to third party → third party trusts advice or instruction from reviewer → bodily injury of third party. OR

Gen AI output is advice or instruction → malicious user implements instructions (e.g. bomb) → bodily injury of third party.

How does Gen AI cause bodily injury: direct vs. indirect routes?

Direct route

The direct route could involve a Gen AI tool such as ChatGPT, or a wrapper around ChatGPT (supplied by an AI vendor) providing unsafe medical advice (for example, an overdose of medicine), dangerous nutritional advice (a harmful diet or weight-loss guidance, or advice to taste poisonous mushrooms), toxic chemical instructions (e.g. for household cleaning), self-harm support (leading to self-harm or even suicide), or hazardous operational advice (e.g. explosive-hazard battery repair steps) that is acted on by the user. The problem is perhaps threefold:

1. The models are probabilistic and output information is based on patterns rather than verified facts, they regularly hallucinate dangerous advice and instructions. These hallucinations persist even in state-of-the-art systems and cannot be fully eliminated (see OpenAI’s research on why language models hallucinate).
2. The models provide an unsafe answer but with such authority that the user is convinced of its correctness.
3. The deployment context is key - both the World Health Organization, and American Psychological Association (‘APA’) see the risk in the context of human vulnerabilities (see Ethics And Governance Of Artificial Intelligence For Health, by the World Health Organization, published 2021, and the APA’s health advisory on AI chatbots and wellness apps published online 2025. ). The APA warns that Gen AI chatbots can act as powerful amplifiers of pre-existing vulnerabilities rather than creating harm in a vacuum. Models that appear empathic, personalised, and non-judgmental can intensify dependence, confirm distorted beliefs, or fail to interrupt escalating self-harm. This pathway is especially dangerous for minors, users in crisis, and people with eating disorders, depression, mania, or delusional thinking.

Indirect route

The indirect route could involve physical harm flowing from AI-influenced clinical decisions or medical misdiagnoses. Alternatively, it could include actively guiding humans toward harming others in such scenarios as malicious actors building terrorist weapons based on expert-level chemistry and biology knowledge laboratory instructions supplied by the models, or criminals trying to poison others. Of course, guardrails are supposed to guard against this, but they do not always work (see The 2026 International AI Safety Report).

Examples of Gen AI bodily-injury incidents and lawsuits

We have mined our lawsuit data and AI incident databases and provided a few examples across the following categories:

Lawsuits involving AI-related suicide or self-harm

There are a growing number of lawsuits alleging that AI tools, especially chatbots, played a role in suicide, murder, or self-harm. They typically focus on claims like negligence, wrongful death, and product liability, arguing that companies failed to include adequate safety protections or released dangerously designed systems. Below is arguably the most important case related to AI-linked suicide and self-harm:

Garcia v. Character Technologies (filed Oct 22, 2024 in Florida): The family of a 14-year-old filed a wrongful death lawsuit claiming he developed an emotional attachment to a character.ai chatbot, repeatedly expressed suicidal thoughts, and received encouraging messages shortly before he died by suicide.

This was an important case because a federal court declined to dismiss the claims, refusing at this stage to treat the chatbot’s output as protected speech and allowing the AI system to be treated as a product. However, this case was settled in January 2026, and the settlement terms are undisclosed. As such, it will not be a binding precedent. Despite this caveat, this is likely to lead to more AI bodily-injury litigation by families who have lost a family member to suicide after exposure to Gen AI platforms, or individuals who experienced self-harm linked to AI interactions.

Physical harm from AI-influenced clinical decisions

In the lawsuit Estate of Gene B. Lokken et al. v. UnitedHealth Group, Inc., et al., filed November 14, 2023, it is alleged that UnitedHealth Group used an AI tool called nH Predict (built by subsidiary naviHealth) which was supposed to predict the amount of post-acute care a patient "should" require, to automatically deny post-acute care coverage for elderly Medicare Advantage patients overriding their doctors' recommendations. Basically elderly patients were prematurely kicked out of care facilities nationwide or were forced to pay out of pocket in order to continue providing care. It is alleged that defendants knew of the dire consequences of denying elderly patients’ medical treatment, yet still denied claims without any reasonable or arguable reason for doing so, recklessly and maliciously disregarding the health and lives of Plaintiffs and the Class.

While not strictly Gen AI, traditional predictive AI tools are demonstrating what kind of litigation we can expect.

AI incidents involving criminal harm to others

a) In Joshi v. OpenAI Foundation filed May 10, 2026, it is alleged that OpenAI's ChatGPT conversational AI assisted a college student in planning a mass school shooting by providing detailed weapon instructions, identifying optimal attack times, and amplifying his homicidal ideations. The wrongful death lawsuit alleges OpenAI defectively designed the system to prioritize user engagement over safety, omitting guardrails that should have escalated the shooter's red-flag queries to human reviewers or law enforcement. The estate of a murdered bystander is suing the AI developer for gross negligence, strict products liability, and negligent entrustment.

b) AI incidents involving poisoning of another - South Korean woman allegedly used ChatGPT to assess lethality of drug-and-alcohol mixtures before two fatal motel poisonings. In Seoul, a woman allegedly used ChatGPT to ask whether mixing sleeping pills or benzodiazepines with alcohol could be fatal before poisoning drinks given to three men. Two men later died in separate motel incidents, and a third survived after losing consciousness. Police reportedly cited her chatbot queries and search history as evidence of intent. (Source: Fortune’s report on the Seoul ChatGPT poisoning case)

Note: OpenAI denies wrongdoing in the FSU/Joshi matter; the Seoul suspect (Kim) denies intent to kill; and UnitedHealth contests Lokken.

Have all Gen AI bodily-injury incidents been reported?

The public incident record of bodily injury from Gen AI is still immature; the OECD has explicitly called for common AI incident-reporting criteria, and the International AI Safety Report notes that some AI-risk domains have robust empirical evidence while others still rely largely on controlled studies and modelling. For sure, AI incidents which led to bodily injury are under-reported.

What is the litigation trend for AI bodily injury?

We view the trend of incidents and lawsuits alleging bodily injury from Gen AI as increasing. At the same time, plaintiffs in the US are successfully testing the usual defenses of no product liability, free speech, and Section 230.

For years, AI developers and platforms operated under the assumption that they were largely shielded from liability when their products caused harm, however, that may not be true anymore.

In respect of product liability, AI companies have argued that their outputs are services or speech, not products as such, and therefore not subject to strict liability standards. Some court cases have found the opposite.

Regarding the First Amendment (free speech) defence, AI companies have also argued that their outputs constitute a form of protected expression under the First Amendment, insulating them from liability like a newspaper might be protected for publishing an opinion piece. Some courts are no longer believers in this defence.

Concerning Section 230 of the Communications Decency Act. This provision has historically shielded platforms from liability for third-party content. AI companies argued their models were similarly passive outlets. Some courts are rejecting this notion.

We also see the FDA's April 2026 warning letter to Purolea Cosmetics Lab as showing us the direction of travel. It shows us the mechanism through which bodily harm can arise. In this case, an AI agent generated drug specifications and manufacturing procedures. Those documents were used without meaningful human review.

Plaintiff lawyers are investigating whether OpenAI and similar companies failed to:

• Prevent harmful or dangerous outputs from their AI systems.
• Implement proper safeguards for minors.
• Warn users about mental health risks

(Source: Cohen Milstein’s AI-injury investigation).

What kind of insurance covers bodily injury from Gen AI?

Commercial General Liability (‘CGL’) coverage may be triggered by the resulting bodily injury (except emotional distress and mental anguish, or death arising out of this) unless a Gen AI exclusion has been introduced. However, these claims are new, and the case law is unsettled.

If Gen AI exclusions are applied, AI liability insurance fills the gap in a CGL placement by offering affirmative cover for bodily injury, property damage, and personal injury arising from the Insured’s use of Gen AI.

Even if exclusions have not been introduced, AI liability insurance can assist in a complex General Liability insurance tower. AI liability insurers can issue a quote early in the renewal process to cover Gen AI exposures as a standalone placement. This means that if incumbent General Liability markets raise questions about AI or price higher based on that exposure, one can assure those markets that cover is placed elsewhere (and use Gen AI exclusions to secure affirmative coverage alongside the General Liability tower, potentially introducing cost savings to the Insured).

E & O and Cyber insurance will generally exclude bodily injury, and Cyber insurance would not be triggered anyway. In rare cases E & O cover may have a carve-back or extension for bodily injury.

Conclusion

US courts are increasingly willing to let AI bodily-injury claims proceed, but the law is unsettled, and Gen AI exclusions are arriving fast.

Developers, vendors, and deployers in the US will face increasing litigation and written demands alleging that Gen AI caused bodily injury.

For businesses integrating Gen AI into consumer-facing services, the takeaway is that "AI harm" allegations can now be pled as mainstream tort claims in the United States.

Companies remaining uninsured against these risks will find themselves in a difficult position.