Key Moments
AI in Healthcare Series: Inside the Rise of AI in Healthcare, Open Evidence and Cyber Risks
Stanford OnlineWant to know something specific about what's covered?
We've already dissected every moment. Ask and we will deliver (with timestamps).
Key Moments
Cyberattacks on healthcare are imminent and sophisticated, posing a greater risk than current defenses can handle, potentially paralyzing care and risking lives.
Key Insights
Nearly two-thirds of physicians are now using tools like Open Evidence, up from 50% just a few weeks prior, indicating rapid adoption.
Nation-states like Iran and North Korea are expected to target healthcare systems, not just with sophisticated AI but also with 'dumb' but effective basic models.
The US government has fragmented cybersecurity responsibility across multiple agencies (Secret Service, FBI, DHS, DOJ), lacking a clear owner and hindering collaboration.
AI adoption is creating a shadow healthcare market where individuals, unable to access traditional care or meet criteria, self-medicate or seek dubious treatments, potentially impacting outcomes.
OpenAI has stated that 1.5 million people weekly ideate on some form of self-harm when interacting with their models, highlighting a critical area for intervention.
Regulatory sandboxes, similar to those used for autonomous vehicles and drone delivery, are proposed for healthcare AI to foster innovation while ensuring transparency and academic study.
Healthcare as a sitting duck for cyberattacks
The healthcare system is currently ill-equipped to defend against increasingly sophisticated cyber threats, a vulnerability exacerbated by the slow adoption of technology and the digitization of records. Despite significant investment in digitizing health records (e.g., $30 billion during the Obama administration), the benefits have largely accrued to payers and systems rather than patients. DJ Patil argues that healthcare should be designated national critical infrastructure, demanding robust defenses. Current hospital systems lack advanced threat detection capabilities like Mythos, and even if they had them, many wouldn't know how to use them. The risk isn't limited to advanced AI; basic, well-executed models from nation-states can cause significant disruption. This inaction leaves systems vulnerable to ransomware and terrorism-level attacks, potentially paralyzing care and endangering lives, especially as budget cuts loom.
Fragmented cybersecurity and the missed opportunity for collaboration
The fragmented nature of cybersecurity responsibility within the US federal government poses a significant challenge. Agencies like the Secret Service, FBI, DHS (CISA), and DOJ each have distinct roles, leading to a lack of unified ownership and hindering crucial collaboration. Patil emphasizes the need to integrate these entities, as external threats and insights are often visible to one agency but not another. Furthermore, classified information shouldn't prevent essential warnings from reaching healthcare systems about impending attacks from entities like North Korea or Iran, as withholding such information endangers lives. The historical lack of a coordinated, government-supported defense strategy for healthcare infrastructure is a critical policy failure.
Increased patient engagement and the rise of AI-powered tools
Despite the cybersecurity concerns, there's a significant positive development in patient empowerment and engagement. Consumers are increasingly taking ownership of their health, utilizing tools like 'Dr. Google' and 'Dr. GPT.' This increased engagement is especially pronounced in areas with limited access to care. Tools like Open Evidence, initially filling a void created by healthcare systems' early reluctance to allow AI use, have seen viral adoption. As of a few weeks ago, two-thirds of all physicians were using Open Evidence, up from 50%. Companies like Consensus are extending this AI-powered access to broader scientific literature. This trend suggests a shift towards greater patient and clinician empowerment through accessible knowledge and potentially more efficient information retrieval, fostering a more engaged healthcare dynamic.
The complex ethical dilemma of AI access for patients
The proliferation of powerful AI tools like ChatGPT for healthcare presents a significant ethical dilemma regarding patient access. Historically, a paternalistic approach in medicine discouraged sharing information directly with patients, fearing radical or ill-informed decisions. However, Patil posits that in an era where advanced technology offers potential solutions to the frustrations of limited access and long wait times in traditional healthcare, the responsibility shifts. The question becomes whether to gatekeep this technology or to open it up, empowering individuals to have more control over their health destiny, especially when traditional care is inaccessible. This moment requires society, technologists, and care providers to reckon with the moral implications of democratizing access to powerful health information.
The emergent 'shadow healthcare' market and unmeasured impacts
The accessibility of AI tools and the limitations of traditional healthcare access are fostering a 'shadow healthcare' market. Individuals are increasingly taking health actions outside standard clinical guidelines, such as using GLP-1s off-label for weight loss or other perceived benefits, or pursuing hormone therapies and other treatments preemptively to avoid future illness or to qualify for coverage. This phenomenon, coupled with potentially dubious 'go peptides' available online, suggests a growing segment of healthcare decision-making occurring independently of clinical oversight. The long-term health metrics of this 'shadow market' are largely unmeasured, creating uncertainty about its true impact on overall public health and potentially leading to unquantifiable health outcomes, both positive and negative.
Struggles with measuring AI's impact on health metrics
A key challenge in assessing the impact of AI in healthcare is identifying definitive metrics that will demonstrably improve. While consumer access to AI tools is widespread, it's unclear if this translates to measurable improvements in health outcomes like diabetes control or blood pressure rates. One perspective suggests that AI could increase knowledge levels, leading to better self-care, provided there is corresponding access to clinical care for necessary interventions like colonoscopies. However, even with AI-driven recommendations, lack of access or affordability can render these insights moot. The potential for AI augmentation in diagnostic testing is significant, but without accessible payment models, its life-saving benefits may remain unrealized for many.
Proposing solutions: Sandboxes, transparency, and ethical frameworks
To navigate the complex landscape of healthcare AI, several interventions are proposed. First, a consortium of model developers, government, and academia could accelerate progress by providing clear recommendations for improving life expectancy outcomes. Second, a move from API to MCIP (Mass Collaboration and Information Platform) could ensure data is used for patient benefit. Payment models need reform to support AI-driven care augmentation. On the ethical front, immediate interventions for AI models interacting with users contemplating self-harm should include directing them to crisis hotlines like 988 or Crisis Text Line (741741). A consortium could also analyze ideation patterns to develop intervention models. Public transparency regarding AI use, data blind spots, and potential harms (like social media's impact on teenage girls) is crucial. The overarching goal is to shape the trajectory of AI in healthcare, fostering innovation through regulatory sandboxes, mandating transparency through reporting requirements on crash data and disengagements, and ensuring academic oversight to learn and scale successful interventions, thereby moving from isolated sparks to an electrified grid of health innovation.
Mentioned in This Episode
●Software & Apps
●Companies
●Organizations
●Drugs & Medications
●Concepts
●People Referenced
Common Questions
Healthcare systems are increasingly vulnerable to cyber attacks due to their slow adoption of technology. These threats range from nation-state actors to simpler, 'dumb' attacks using basic AI models, capable of causing widespread disruption and paralysis of care.
Topics
Mentioned in this video
Mentioned in reference to his house in the Bay Area.
CEO of Seattle Children's, formerly a proponent of government assistance for healthcare cybersecurity.
A person DJ takes a class with, who was discussed in relation to healthcare metrics and AI.
Mentioned as a proponent of public health thinking.
A company founded by DJ.
A health insurance company where DJ is involved, utilizing AI.
A social media platform where people find health information and make decisions.
A social media platform where people find health information and make decisions.
The company that released data on AI's interaction with self-harm queries and is being pushed for more transparency by California.
A company whose reduction in crash rates is cited as a positive example of transparency and progress in autonomous vehicles.
Mentioned as a company that entered the market, alongside Waymo's progress.
A company that faced backlash for prescribing medicines in Utah without a proper framework.
A project mentioned in the context of cybersecurity and hospital vulnerabilities.
A powerful model that can hack common software, mentioned in the context of potential healthcare vulnerabilities.
A company that provides AI tools similar to Open Evidence for general scientific research.
An AI tool specifically for clinicians, requiring NPI verification.
A large language model discussed as the largest health application used by consumers.
An AI model mentioned as a potential source of health advice, alongside ChatGPT.
A crisis support service accessible via text, recommended for AI systems to direct users to.
The administration during which the digitization of the health system was initiated with significant taxpayer funding.
A children's hospital where Chris Longhurst is CEO.
A tool discussed as being useful for physicians, with a high adoption rate.
Mentioned as one of the agencies with competing areas in cybersecurity.
Mentioned as one of the agencies with competing areas in cybersecurity.
Department of Homeland Security; mentioned as one of the agencies with competing areas in cybersecurity.
Cybersecurity and Infrastructure Security Agency; mentioned as one of the agencies with competing areas in cybersecurity.
Department of Justice; mentioned as having separate portions of cybersecurity responsibilities.
An organization trying to advance healthcare innovation.
The administration that previously focused on health AI assurance labs.
The current administration which has a more 'all gas, no brakes' approach to AI, but is beginning to show concern.
Referred to as 'the big beautiful bill', it is mentioned as causing budget cuts and downstream impacts on health systems.
Affordable Care Act; mentioned as a factor affecting access to care and ability to afford services like colonoscopies.
Current Procedural Terminology codes, mentioned in the context of potential payment model reforms for AI in healthcare.
More from Stanford Online
View all 78 summaries
42 minStanford CS153 Frontier Systems | Scale, AGI, and the Future of Everything
60 minStanford CS547 HCI Seminar | Spring 2026 | The Modern Motivators of Play
72 minStanford CS336 Language Modeling from Scratch | Spring 2026 | Guest Lecture: Dan Fu
50 minStanford MS&E435 Economics of the AI Supercycle | Spring 2026 | Applications, Applied AI
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free