--- title: "How healthcare's AI tools turn into liability claims (Webinar Replay)" url: "https://www.testudo.co/webinar/ai-in-the-healthcare-industry" description: "A briefing for brokers and risk leaders on how healthcare providers adopted generative AI, the privacy lawsuits and incidents that followed, and what to ask clients about governance and consent." site: "Testudo" --- # How healthcare's AI tools turn into liability claims. ## Chapters ยท 12 1. [0:00 Introduction & AI in Healthcare Overview](https://www.youtube.com/watch?v=YRfUJ4eTnYc) 2. [1:26 Background on Generative & Agentic AI in Healthcare](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=86s) 3. [3:46 AMA Report: Physician Concerns & Training Gaps](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=226s) 4. [6:28 Testudo Lawsuit Database & Deployer vs. Vendor Liability](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=388s) 5. [7:43 Case Study: Washington v. Sutter Health and MemorialCare](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=463s) 6. [12:04 Case Study: Tate v. Chemed Corp / VITAS Healthcare Corporation](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=724s) 7. [16:14 Testudo Intelligence Reports & Marketing Exposures](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=974s) 8. [17:20 Incident Database: Shadow AI, Clinician Pushback & VA OIG Report](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1040s) 9. [19:47 Incident Case Study: Loma Linda University Health PHI Exposure](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1187s) 10. [21:16 Coverage Comparisons: Medical Malpractice & GL vs. Testudo](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1276s) 11. [22:48 Healthcare AI Claim Scenarios: Bodily Injury, Property Damage & Financial Loss](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1368s) 12. [27:18 Closing Remarks & Broker Resources](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1638s) ## About this session Healthcare organizations are using AI scribes, chart summaries, and patient communication tools across clinical and administrative work. Arden Hovermale starts with the American Medical Association's 2026 survey of 1,692 physicians, which found over 80% using AI at work, a quarter with no training, and patient privacy as the main worry. She then goes through two lawsuits from Testudo's AI lawsuit database, Washington v. Sutter Health and MemorialCare and Tate v. Chemed Corp. and VITAS Healthcare, and the incident reports on shadow AI, clinician pushback, the VA Inspector General, and Loma Linda University Health. The session closes with claim scenarios that show where medical malpractice and general liability forms leave gaps, and how consent rules vary by state. ## Key takeaways - Over 80% of physicians surveyed by the AMA now use AI at work, yet a quarter have had no AI training and patient privacy is the one area where most expect harm. - AI vendors usually cap their own liability for what their tools produce, so the hospital or practice that deploys the tool absorbs it. - The scribe and call-recording lawsuits turn on whether the provider had an AI usage policy, a system-wide way to collect patient consent, and a record of telling patients the tool was in use. - Plaintiffs quote a deployer's own marketing about its AI partnerships to argue it should have taken more care, so how a client talks about AI in public matters. - Staff using unapproved AI tools, as in the Loma Linda incident, is a governance gap to ask every healthcare client about. ## Transcript Search the transcript ### 0:00 Introduction & AI in Healthcare Overview - [0:00](https://www.youtube.com/watch?v=YRfUJ4eTnYc) All right, hello everyone. Thank you for joining and welcome to another AI liability session with Testudo, where today we'll focus on the healthcare industry's adoption of AI tools. We welcome all of our attendees today and want to say thank you for your time. Now with your help we can raise questions around emerging risks with our clients and ensure that cutting-edge technology can be implemented safely while still delivering growing benefits to physicians and staff in the healthcare space. - [0:26](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=26s) So some topics that we'll address today include the industry's adoption of AI tools over the past few years, an analysis of litigation and incident examples from our database here at Testudo so we can assess how third-party harms arise when healthcare entities deploy these tools and where liability risk emerges, and then some guidance for brokers so that we can identify high-risk use cases and prudent governance controls. - [0:48](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=48s) So to begin we'll go through a fantastic report by the American Medical Association from earlier this year on industry adoption of AI tools and with this context we can then get into some AI litigation incidents from our own database so that we can track the validity of industry adoption and map where the use of these tools materializes as harms. We'll finish with an overview of resources for brokers to isolate those risks and assist their clients. - [1:12](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=72s) So we hope that our session today delivers intelligence that you can rely on to navigate conversations around AI liability risk and affirm that the resources we offer to assist you in these conversations can be accessed 24/7 with our team. So the American Medical Association's Center for Digital Health and AI published in March this year a report on the industry's adoption of AI tools both by patients and clinicians. ### 1:26 Background on Generative & Agentic AI in Healthcare - [1:37](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=97s) So the survey was conducted between January to February this year over total of 1,692 physicians to survey changes from previous reports done since 2023. Now the report notes that over 80% of physician respondents currently use AI in a professional context which is double the number first reported in 2023. There's a general excitement in use cases for documentation and summarization tools. So, this means that the use of AI for summaries of research and standards of care has increased dramatically since the last report with nearly 26% more physicians using these tools in professional workflows since 2024. - [2:15](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=135s) Now, physicians perceive AI tools to bring positive benefits to their workload. The main driver of physician use of AI, according to 70% respondents, is the opportunity to automate the clinical and administrative tasks which contribute to physician burnout. So, as a result, we see that the most popular use cases are concentrated in summaries of medical research and standards of care, creation of discharge instructions, care plans, and or progress notes, documentation of billing codes, medical charts, visit notes, and the generation of chart summaries. - [2:45](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=165s) So, clearly there's a benefit to relying on these tools for heavily administrative tasks. For example, physicians who must conduct manual summaries of patient visits see their capacity for meeting face-to-face diminish by the time required to draft summaries. So, AI scribe tools used for both summarizing written content and transcribing real-time conversation into physician notes has delivered time back to those physicians. - [3:07](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=187s) Vincent Liu is the chief data officer at Kaiser Permanente Medical Group. He states that both doctors and patients highly value face-to-face contact during a visit, and the AI scribe tool supports that. Finding the capacity for physicians to maximize face time with patients has been unlocked by AI scribe tools. But interestingly, vendors of AI scribe tools have begun to differentiate their user base beyond just physicians to now include nurses and other members of staff based on the perceived benefits and time given back to healthcare staff. - [3:35](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=215s) This is critical to understand given estimates that the US medical AI scribe market is anticipated to reach a value of nearly 3 billion dollars by 2033. Now, in analysis of where physicians use AI scribe tools us most, AI-generated reports deliver the most time back to physicians and staff. The AMA report highlights that where AI tools have delivered the most assistance and been most helpful have been for work efficiency, diagnostic ability, cognitive overload, clinical outcomes, patient convenience, and stress and burnout. ### 3:46 AMA Report: Physician Concerns & Training Gaps - [4:08](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=248s) The only factor where most respondents anticipate AI to cause harm, however, is in regards to patient privacy. So, it's interesting that Liu mentions how doctors and patients value maximizing face-to-face time, which could be done through the use of an AI scribe, as industry adoption has raised some concerns about physicians relying on these tools. While institutions are contemplating how to vet and roll out AI tools, their physicians and staff may already be using them. And we'll return to this concern manifested as an AI incident momentarily. - [4:36](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=276s) Most concerns by physicians around patient privacy involve whether the tool has been sponsored or deployed by the institution. So, 71% of physicians are understandably worried about patient privacy regarding AI that the institution has not officially deployed, but which is being used and relied upon by staff. But a significant 42% of physicians remain concerned even when their own institution has officially deployed the tool. - [5:01](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=301s) Now, the factors influencing this confidence and concern come directly from the institution's stance on their own AI deployment. So, what increases confidence or lessens the worries of these professionals is when their own practice or hospital and their EHR vendor assures them on data privacy and security. However, what does not push physicians towards confidence is when their practice leaders recommend the tool. So, clearly there are more trustworthy sources of truth in terms of building confidence in AI systems, and these sources of truth tend to come from experts vetting and handling sensitive patient data. - [5:32](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=332s) And then downstream of this discussion comes the liability frameworks for institutions which deploy these tools. So, physician respondents to the survey are actively interested in learning who holds legal responsibility and accountability in cases where patient privacy, safety, or clinical efficacy are compromised. The AMA report notes that "establishing clear liability frameworks was most frequently selected as the most important regulatory action that would increase trust and likelihood of adoption of AI tools." - [6:00](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=360s) Now, what makes this even more interesting is the figure that 1/4 of physicians surveyed had received no AI training from any source. 92% expressed interest in more training being delivered, leading us to understand that a chronic lack of training, coupled with rapid adoption and expansion of capabilities of AI tools in clinical settings, has enticed physicians enough with perceived benefits, but comes in tandem with significant concerns around patient privacy and where liability falls in the event that something goes wrong. ### 6:28 Testudo Lawsuit Database & Deployer vs. Vendor Liability - [6:30](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=390s) Now, as we all know, Testudo tracks real-time emerging signals of AI-related incidents and AI-related litigation to assess where harms occur when businesses deploy AI tools and how those harms translate into legal theories. So, by doing so, our underwriting seeks to understand where and how AI liability emerges. Our understanding is that typically AI technology vendors limit their liability for AI-generated outputs of the systems which they sell. And that could be the response of chatbot or an AI-generated patient chart summary or the AI scribe-generated transcription of a physician-patient meeting. - [7:03](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=423s) And then the deployer of this tool, which is the business, the clinical practice or the hospital, they absorb the liability for these outputs. So, when we look to examples of how this plays out in litigation, there are consistent themes raised in complaint documents, and plaintiffs' lawyers are wording complaints in increasingly sophisticated ways. So, they're actually wording the source of harm to include the AI-generated output and how that output was put into commercial use where that third party interacted with it. - [7:27](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=447s) And they even leverage the deployer's marketing and advertising of their use of AI against them. So, we'll see these themes in two lawsuits from this year, Washington versus Sutter Health and Memorial Care and Tate versus Chemed Corp or VITAS Healthcare Corporation. So, in Washington versus Sutter Health and Memorial Care, the class action raised earlier this year alleges that the two health care institutions deployed AI scribe tools sold by Abridge, which is popular vendor of tools which assist physicians with administrative tasks. So, the tool can summarize those physician-patient conversations and generate summaries. ### 7:43 Case Study: Washington v. Sutter Health and MemorialCare - [8:04](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=484s) The class action details how the use of this tool, while in the hands of the health care institution, forms the nexus of harm, being that the tool "captures the substance of what both the patient and physician say, converts the recorded speech into text, and applies artificial intelligence models to generate structured clinical documentation." So, the argument for patient harm alleges that "despite the sensitive nature of these communications and the legal protections governing them, defendant failed to implement a standardized or system-wide procedure to obtain valid all-party consent before recording physician-patient conversations using the Abridge system." - [8:40](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=520s) Now, this may seem like a typical privacy lawsuit, and being based in California, there is naturally a large corpus of privacy statutes, as well as the Federal Wiretapping Act that could be triggered. And so, it may not seem like a brand new risk for businesses to watch out for, but given the use of tools which are now generating summaries of these recorded conversations, the lawsuit frames this in a clever way to allege the institution's liability. - [9:02](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=542s) So, it alleges that defendants permitted the recordings and transcripts generated through the Abridge system to be sent to third-party infrastructure for processing outside the clinical setting. So, the harm comes from the conversation summaries being improperly generated and disseminated while in the hands of the health care institution. Meaning the institution is the entity which allegedly permitted this generation and transmission without collecting patient consent or disclosing to patients that this tool was being used. - [9:27](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=567s) And thus, the AI output was allegedly generated and used without reasonable care given to the usage processes by the institutions. So, these allegations of why the health care institution is liable for AI generated outputs causing third-party harm are raised in an interesting manner in the common questions of law and fact where plaintiffs ask, among other questions, whether defendants had a policy or practice of using Abridge or similar technology to record patient conversations without obtaining all-party consent. - [9:54](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=594s) And then whether defendants failed to implement any system-wide protocol reasonably designed to ensure that patient consent was obtained prior to recording. And whether defendants sought or obtained prior consent, express or otherwise, from plaintiffs in the class. So, these questions raise three core considerations for any commercial entity deploying similar tools. They ask whether the business had a policy or practice in place for the use of this technology, whether their system-wide protocol was reasonably designed to collect consent, and whether consent was sought in express terms. - [10:24](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=624s) These considerations are crucial as they indicate where AI liability litigation is moving. The healthcare institution's defense now lies in whether they had a clinical AI usage policy, procedures on collecting patient consent, and whether their physician protocol was reasonably designed and not negligent in navigating patient disclosure and consent so that physicians could use the tool. - [10:46](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=646s) So, as many of our brokers will recognize, these are factors that we ask for on our application form. An AI usage policy, third-party disclosures, and PII and PHI processing consent collection. So, they're all exclusively employer-specific exposures if unaddressed. If the AI vendor limits their liability for the outputs of their system causing harm, the employer must now absorb the responsibility of generating these outputs in a governed and reasonable manner, which boils down to procedures for employees on how to use the tool, demonstrated care taken to obtain consent, and disclosing to third parties when AI tools are being used in processing their PHI. - [11:22](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=682s) So, when we go back to the AMA's report on physician use of AI tools, the figures that now concern us are that 1/4 of physicians have received no AI training, and almost half remain concerned about patient privacy even when the AI tool has been deployed and sponsored by their own institution. So, does this mean that the bulk of physicians are not being instructed by any protocol on when to use AI scribes? And are there policies in place, but perhaps left unenforced? - [11:46](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=706s) Are physicians being instructed to disclose to patients and collect consent when they use the tools? Because when 92% of physicians report that they would prefer more AI training, how many healthcare institutions will be able to defend themselves with evidence that they took reasonable care to the deployment of AI tools among their physicians? Now, the second lawsuit we'll look at is Tate versus Chemed Corp or VITAS Healthcare Corporation. ### 12:04 Case Study: Tate v. Chemed Corp / VITAS Healthcare Corporation - [12:10](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=730s) So, in short, this is another case where a healthcare institution allegedly recorded physician-patient conversations without collecting consent. And these transcripts and generated summaries were allegedly sent to a third party, being the technology vendor, Invoca, without patient's express consent. However, this case is interesting for how the allegations are raised to argue that VITAS did not take reasonable care to deploy this tool. - [12:33](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=753s) So, the lawsuit alleges that VITAS used a tool by Invoca, which "provides a searchable database of every call transcribed with best-in-class accuracy by generative AI, and allows for the review of transcript for unique insights into consumers, products, call experience, and more." So, this tool was allegedly used by VITAS call centers without obtaining third-party consent to record and transcribe the conversations, allegedly striking up against the California Invasion of Privacy Act and the Federal Wiretapping Act. - [13:01](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=781s) It's not entirely surprising given the use of new technology for transcribing conversations, and as we saw in Washington versus Sutter Health, the defense of the employer here now rests in whether they had procedures in place for collecting third-party consent. But, what interests us here is how plaintiffs raise the defendant's involvement in facilitating this harm. So, the lawsuit raises that VITAS, the nation's largest hospice and palliative care provider, is an Invoca customer. - [13:25](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=805s) And citing their website, VITAS receives over 90,000 annual calls and growing in its three call centers throughout the nation. They even cite that VITAS can adjust the content that it delivers based on signals ascertained from AI-generated analytics about these calls. And the source used here is a YouTube video by Invoca on VITAS Healthcare between the Invoca CEO and VITAS Healthcare's Chief Marketing Officer. - [13:48](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=828s) Their other source is Invoca's website at invoca.com/customers/vitashealthcare. This VITAS-specific link now anonymizes their customers. But these sources, which are co-marketed by Invoca and VITAS, are used to frame the partnership as well-formed. And so the now-removed VITAS customer page on Invoca's website is also used as a source in the lawsuit's detailing of how VITAS uses Invoca to record calls so that VITAS has a clear picture of call activity, including what families and referral sources commonly say. - [14:19](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=859s) So, taking a look at these sources, the picture they paint of the defendant is intended to highlight their sophisticated marketing and partnership ventures. But failure to responsibly govern these AI tools in settings with vulnerable third parties. So it frames a push in marketing VITAS partnership with Invoca to frame the company as a digital-first organization, which wants to "give patients and families the ability to do as much through digital as possible, but also recognizing that hospice is just a tremendously emotional decision." - [14:47](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=887s) And this comes from the VITAS CMO, who goes on to say, "For me, it's sort of an interesting interplay of how you do the right things through digital and how you do the right things through the human touch." Now this paints the company as a, you know, this is a figure who would be highly aware of how to do the right things with technology's assistance, which sets a much higher bar for taking responsible and reasonable care to ensure that the use of these tools is well-governed and deployed. - [15:11](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=911s) Now, this is a material concern for traditional businesses where we anticipate their risk managers may not be considering this enough. So to compete in the the industry with the narrative of serving their customers through excessive digital means comes with its own risks. The marketing of these services, once harm does indeed occur, can be used against these institutions. So, this massive push in digitizing call transcriptions in order to integrate generative AI has led to alleged privacy invasions. And the argument that because the tool is in the hands of the healthcare organization, it is their fault, not the vendor's faulty design, that caused these harms. - [15:48](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=948s) And watching their marketing content post-lawsuit gives the healthcare technology provider partnership and intentional feel, which does not help the healthcare institution in the event that this AI tool, used in a high-risk setting, led to third-party harm. And that they now must prove the reasonable care taken to operate such tools. The public perception of a business now using generative AI tools would be more impactful in subsequent legal action than anyone can currently anticipate. ### 16:14 Testudo Intelligence Reports & Marketing Exposures - [16:14](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=974s) So, this is why Testudo offers intelligence reports to our broker partners. And these reports identify how companies have publicly marketed their use of AI tools, and allows the broker and their client to consider that in the event that their AI system causes third-party harm, how that company will be perceived by the public. When lawsuits use these public sources to raise questions around whether reasonable care was taken to train employees on how to use AI tools, and whether patient consent was a deep consideration in AI deployment, it puts pressure on companies around how they will defend themselves if their tools cause harm. - [16:46](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1006s) So, companies will want to defend their position with evidence of having anticipated the need for robust controls in place, thereby safeguarding the deployment of generative AI with, for example, human review and consent collection. So, we actually ran an intelligence report on VITAS, which a picture of is here on the slide, which surfaced Invoca's marketing web pages, as well as those by other technology vendors that partner with VITAS. So, this report frames a company's public reputation under the lens of its AI deployment exposure. - [17:14](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1034s) So, in the event of third-party harm, it's a good question to ask. How would your clients look? Now, at Testudo, we also collate AI-related incidents, which may contribute to material AI litigation down the line. So, we're going to go through a few sources commenting on the use and impact of AI-scribe tools in the healthcare industry and actively think about how public perception of these tools could contribute to legal arguments if these tools cause harm. ### 17:20 Incident Database: Shadow AI, Clinician Pushback & VA OIG Report - [17:40](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1060s) So, the Markup's article raises concerns about whether patients knew that their medical providers may be recording their visit. It states that in 2024, Kaiser Permanente announced the rollout of Abridge, describing it as designed to help clinicians, including mental health providers, securely capture clinical notes during patient visits. A licensed clinical social worker speaks on this and attests that she chose not to use the platform because in early meetings internally, she and others raised questions about patient privacy and HIPAA compliance and technology safeguards, but they were met with "empty assurances" from leadership that the institution was compliant. - [18:16](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1096s) So, why this matters for these workers is the pressure put on them to meet with more patients in recent years than ever before as "refusing to manage the increased patient volume can be treated as a failure to meet job expectations." But, management frequently recommends the use of Abridge to save time, raising ethical concerns over pressuring staff to use the software. What further complicates patients' role in these situations is that patients often have little visibility into the arrangements between vendors and the healthcare employer. - [18:45](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1125s) While they have the right to refuse recordings, the opt-out process is not always clearly presented. In another incident, the Department of Veterans Affairs Inspector General noted in June this year that they allow their employees to use VA GPT and Microsoft Copilot in veteran healthcare operations. But, neither are designed for clinical use and the VA "does not centrally curate or evaluate prompts nor are the generated output that could be applied to clinical decision-making." - [19:13](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1153s) So, the agency's Office of Inspector General noted that this "lack of appropriate oversight or safeguards is creating risks for patient safety and limiting the ability to monitor AI chat tool related errors." So, this report notes specifically that the VA's use of this ambient scribe tool is a high-impact use case requiring human oversight and has stressed recommendations of addressing use and oversight of GenAI chat tools, evaluating the tools as high-impact and requiring safeguards, and integrating risk monitoring into patient safety programs. - [19:46](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1186s) And finally, an interesting update from July this year when Loma Linda University Health notified patients after a file containing protected health information was inadvertently uploaded to an external AI platform. The exposed file may have included "patients' medical record numbers, dates of birth, and limited clinical information tied to orthopedic care." But in a news release, Loma Linda confirmed that social security numbers, financial information, insurance information, and complete treatment records were not involved. ### 19:47 Incident Case Study: Loma Linda University Health PHI Exposure - [20:13](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1213s) Now, what this raises are questions about shadow AI or the use of unapproved AI tools by enterprise staff. So, when a business deploys an enterprise-grade chatbot, it is typically agreed between themselves and their vendor that information uploaded will not be used to train that AI tool or that model, and that data will only be stored on vendor servers for a specific amount of time. Now, these agreements are why businesses are able to upload confidential information into AI tools and benefit from the use of these tools securely. - [20:41](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1241s) But for some businesses who've not yet deployed a multi-use chatbot, or in the instance of health care entities, an AI scribe tool, employees may take to using their own personal non-enterprise AI tools to ease their workload. And so, while this is not confirmed in the Loma Linda incident, that the institution had to notify affected patients after it was discovered that confidential information was uploaded to an external AI platform suggests that external here means an AI system not deployed to staff by Loma Linda and therefore not secure and vetted for employee use with confidential patient information. ### 21:16 Coverage Comparisons: Medical Malpractice & GL vs. Testudo - [21:18](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1278s) So we've given a lot of context as to incidents and harms occurring in the healthcare industry as a result of physicians and staff relying on AI tools. Some harms we haven't even touched on yet include bodily injury, personal injury, and third-party financial loss as a result of relying on AI generated outputs. Beyond the use of AI scribe tools, there are medical tools available to assist in diagnostics, tools that can take a patient chart and summarize key points for the next nurse or personal shift, and even tools that generate care plans for patients being discharged. - [21:49](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1309s) One of the most important conversations that we have with our brokers is coverage comparison so that brokers can ensure their clients that a new liability gap created by the use of these tools is indeed covered affirmatively. So we'll get into two key areas where coverage comparison is essential based on coverage that healthcare institutions currently purchase, which is medical malpractice and cyber insurance. To begin with, MedMal. - [22:12](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1332s) Typical MedMal forms address GL including personal injuries, property damage, and advertising injury, and medical professional liability responding to medical incidents which give rise to personal injury of these patients. So because Testudo's form schedules the AI system and use case as deployed by the insured, regardless of whether, for instance, bodily injury arises under a general liability or a professional healthcare services claim, if it arose from the named insured's use of a scheduled AI system for its scheduled use case, it would be covered by Testudo. - [22:42](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1362s) Now this removes coverage ambiguity, particularly because our form is on a primary basis. To get into a few loss scenarios, for bodily injury, if a healthcare professional uses an AI meeting transcription tool to transcribe their client meetings, at the end of a meeting, it may generate a summary. And if the physician reviews this before it goes into patient's chart, they could fail to flag a key error in that summary. ### 22:48 Healthcare AI Claim Scenarios: Bodily Injury, Property Damage & Financial Loss - [23:05](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1385s) And another physician could make a medical decision based on that hallucinated AI output, resulting in patient bodily injury. So, a med mal form could treat this as a medical professional liability claim. As the patient's failure to identify an error in an AI-generated summary, followed by another physician relying on that summary, it could fall within the professional liability framework. But, the med mal form does not have language around the intervention of an AI output in that decision. - [23:29](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1409s) So, given our understanding that may not be properly trained on how to use these tools for their work, it may be difficult for the institution to prove that it was in their station of the physician's professional work to identify errors and correct them. Testudo's form only requires that bodily injury be sustained by a third party to whom you owe a duty of care arising from your deployment of a generative AI system. - [23:52](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1432s) Regarding property damage, a home health care provider uses generative AI to create equipment handling and home care instructions for clinicians who are seeing patients in their homes. Now, the AI instructions contain an error. So, when a clinician follows the AI-generated instructions, they accidentally physically damage the patient's property. Now, the med mal form may cover property damage under GL, but it excludes property damage resulting from professional health care services. And it's medical professional liability responds to injury arising from a medical incident only. - [24:21](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1461s) So, Testudo's form only requires that property damage be suffered by a third party to whom you owe a duty of care arising from your deployment of a generative AI system. Now, on third party financial loss, let's say a hospital uses a generative AI tool to create public-facing website content describing the capabilities of one of its specialist programs. And this AI-generated content incorrectly states that the hospital offers a particular procedure and has specialist expertise that does actually not actually doesn't possess in real life. - [24:49](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1489s) And so, if a patient were to rely on those representations when they choose a hospital, they may incur a substantial travel, consultation, and related expenses, and forgo treatment elsewhere. And after learning that the hospital cannot provide this represented service, the patient may bring a claim against the hospital alleging negligent misrepresentation, and seeking recovery of those financial losses. Negligent misrepresentation, the hospital possesses clinical capabilities it does not have, where the claimant seeks purely economic loss, also does not fit comfortably within the med mal advertising injury definition. So Testudo's form provides an affirmative route to coverage for the financial loss arising from an AI generated misstatement about the hospital's capabilities. - [25:30](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1530s) Regarding cyber comparisons, we can go back to the Sutter Health lawsuit. So, a few details from this complaint included allegations that this AI generated summary containing PHI held by Sutter Health was unlawfully held due to questions over whether they actually collected patient consent when using an AI scribe. Now, to be clear, our form would not cover wrongful or unlawful collection, but we would affirmatively respond to incidents where accidental or inadvertent disclosure of lawfully held data takes place. - [25:57](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1557s) Another important distinction here is to look at what lawfully held actually means in each state given their emerging AI regulation. So, 13 states require all-party consent, but all other states only require one-party consent. So, a physician using a transcription tool could actually qualify as one party consenting to this. And practically speaking, a recorded notice and agreement script at the start of each visit with this recorded conversation, such as "this visit will be recorded and transcribed by an AI tool, is that okay?" That can satisfy most states on this requirement. - [26:30](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1590s) And now finally, Testudo offers AI regulatory violation cover. So, state AI regulation frequently targets healthcare as a regulated industry with high exposure to vulnerable third parties. So, regulation increasingly mandates that regulated industry players disclose to third parties when their AI system is used to process protected information, when they are interacting with an AI tool, and when an AI tool is being used to form part of a consequential decision, which can mean third-party access to health care services. - [26:58](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1618s) If the named insured were to fall afoul of a relevant AI regulatory requirement, and an authority such as the Attorney General were to bring a fine, penalty, or a requirement of that named insured to comply with this regulation, Testudo's form could cover the costs of cooperating with such an authority, and fines and penalties were insurable. And with that, we'll close out today's webinar. So, as mentioned throughout presentation, if you are a broker interested in providing this coverage to your clients, let us help you identify their exposure and where affirmative AI coverage can help their business. ### 27:18 Closing Remarks & Broker Resources - [27:29](https://www.youtube.com/watch?v=YRfUJ4eTnYc&t=1649s) We are more than happy to run intelligence reports on your clients, so that you can go into these conversations prepared, and with the tools necessary to enable your clients as modern enterprises to use AI while minimizing the downside. So, please be in touch with any questions, and we look forward to working with you. Thank you. Recorded October 14, 2026 Length 27:47 min Hosted by Arden Hovermale Underwriter of AI Risks The next live session is October 14. [Save your seat](https://www.testudo.co/webinar) [Watch on YouTube](https://www.youtube.com/watch?v=YRfUJ4eTnYc) --- ## More from Testudo - [Site index for agents](https://www.testudo.co/llms.txt) - [How to use Testudo (agent instructions)](https://www.testudo.co/agents.md) - [Coverage summary](https://www.testudo.co/coverage.md) - [Sitemap](https://www.testudo.co/sitemap.xml)