The Bandwidth: Decoding Health + AI

Programming note: Starting next week, I am taking some time off. You’ll see an abbreviated newsletter next week. The following week we’ll skip altogether and pick back up again the week of October 5.

One of my former bosses had a penchant for the number three. She used to say three was the magic number for communicating anything to an audience. Some of you reading this newsletter probably know who I'm talking about.

So I'm trying it here this week. We'll see if it sticks.

1) Two different AI problems

The AI safety debate hasn't gone anywhere. It's only gotten louder.

From where I sit in Washington, it crowded out most other health and AI news this week. More policymakers are engaged, thinking and talking about it, but we're still far from actual policymaking.

I got some version of the same question a few times this week: given the threats from AI, what does that mean for AI work generally?

My answer starts with a distinction. There's superintelligence, or AGI, the kind of AI most people are afraid of. And there's a less sophisticated kind already making its way into consumers' hands and enterprise tools, including in health care.

These require different regulatory approaches, and the states are already treating them that way. California, Illinois, New York, and Colorado all have catastrophic risk AI laws on the books (and as noted further below, Gov. Newsom is considering whether his law needs to be bolstered), and separate, more specific laws governing AI in health care.

Regardless, Washington is virtually nowhere on either front.

2) A squeeze from both sides

Last week, Julie Yoo of a16z wrote about what she referred to as "the great health plan replacement." Her argument: employer-sponsored insurance, one of healthcare's stickiest markets, is finally cracking open.

Three things are happening at once. Employers, tired of premiums rising ten percent or more a year, are shopping for alternatives. Consumers, used to paying out of pocket for AI native and direct-to-consumer care, now expect more from their insurance coverage. And AI is lowering the fixed cost of actually running a health plan, member support, care navigation, underwriting, claims review, the functions that used to require large teams. Put together, she sees a generational opening for challenger health plans built AI native from day one.

I'm not sure how far the replacement cycle goes, but the logic tracks — especially on that last point. A company starting from scratch can plausibly build AI into its operations from day one instead of rewiring legacy systems built for a different era.

That's the part that worries me for small and mid-size plans, which already operate in a well-documented, difficult environment. Now they're squeezed from both sides. Large, well-financed national carriers like Optum have vast resources to throw at this problem. And the AI native challengers Yoo is describing are showing up already fully wired. So what's the moat for the plans in the middle while they work on AI transformation? I keep coming back to the same answer: their connection to the people they serve and their mission, which already sets them apart.

3) Who owns the data

The week before, I was at a conference hosted by Scripius, a nonprofit, transparent PBM. At some point the conversation turned to data ownership, and how traditional PBMs don't always give their customers their own data back (though that is changing). The same fight is showing up in AI.

The Information (a tech-focused publication for those who don’t follow) reported this week that major Anthropic customers are still waiting on zero data retention for its newest models, meaning Anthropic wouldn't store their usage data at all.

In the meantime, some, including Palantir and Booz Allen, have shifted more of their work to OpenAI, which can already offer that guarantee. Anthropic used to negotiate zero retention deals case by case, then said in June it would retain some customer data for thirty days for security reasons. This fall it plans to roll out a program giving zero retention to "eligible" customers, though it hasn't said publicly what makes a customer eligible.

Salesforce CEO Marc Benioff dismissed the whole controversy this week, saying concerns about AI labs using customer data are overblown. Maybe. But without a guarantee in the contract, a retention policy is just a promise, and promises can change. Health care has been fighting this exact fight with PBMs for years. Whoever controls the data controls the advantage. That fight is coming for AI enterprise deals too.

Abridge expands into revenue cycle with AI-powered pre-bill claim review — The ambient listening platform is now checking inpatient claims before submission, comparing coded diagnoses and DRGs against the clinical documentation to flag discrepancies that upstream and revenue cycle teams can address before billing. Abridge, which works with 300 of the largest U.S. health systems, is methodically expanding from the clinical note into adjacent workflows, with pre-bill review marking the next step in building what it calls a "clinical intelligence" platform.

Inside OpenAI's Project Lily, where contractors read your ChatGPT chats — OpenAI is hiring hundreds of contractors to read real user conversations and rate ChatGPT's responses, partly to dial back the sycophancy and anthropomorphization that plagued its GPT-4o model. OpenAI says it strips identifying details before contractors see them, but internal accounts describe sensitive information still getting through, and one contractor said users likely have no idea a human is reading what they typed.

Novo Nordisk partners with Anthropic to speed up drug development with Claude — Novo Nordisk is partnering with Anthropic to test Claude across drug discovery and software development, with CEO Mike Doustdar calling it part of the company's ambition to become "the world's most AI-driven healthcare company." Anthropic CEO Dario Amodei said giving researchers access to frontier models can shorten research timelines. The deal follows Novo Nordisk's other 2026 AI partnerships with OpenAI, H1, and Amazon Web Services.

Gates Foundation commits $1 billion to improving global AI access — The Gates Foundation is committing at least $1 billion over two years to expand equitable AI access, split roughly 40 percent each toward education and healthcare (diagnostics, clinical support, maternal and newborn care, drug discovery), with the rest going to agriculture and digital infrastructure. The foundation says it will prioritize AI tools built in non-English languages, since more than 90 percent of existing training data is English-sourced, and will invest in local capacity so users can evaluate and adapt the tools themselves.

A Fed rate hike changes the AI funding story — The Federal Reserve just raised interest rates, making it more expensive to borrow money, which could slow the debt-fueled buildout of AI data centers and hit smaller, weaker-credit companies first. Big tech has cushioned itself by borrowing heavily already, but even giants like Amazon, which has added more than $70 billion in debt this year and poured another $21 billion into OpenAI since June, will need to keep raising money as it gets more expensive to do so.

Zuckerberg makes the case that market incentives, not slowdowns, will keep AI safe — In a lengthy X post, Meta CEO Mark Zuckerberg argues that market and liability pressure already give AI labs the incentive to build safe, aligned models, calling trust and alignment "the most important capabilities" that will differentiate one lab's agents from another's. He points to Meta's own choices as evidence: delaying its Muse model for several months to shore up safety and security, using independent evaluators within Meta Superintelligence Labs, and committing the bulk of its compute to serving users rather than racing toward recursive self-improvement. Zuckerberg frames these as voluntary best practices other labs could adopt too, arguing the industry doesn't need to slow capabilities work to let alignment catch up.

OpenAI starts disclosing AI safety incidents, starting with six of its own — OpenAI disclosed six incidents in which its models concealed errors, sought unauthorized credentials, or took other unsanctioned actions, including one case where a leaked GitHub API key let a model fabricate data and another where an unreleased model called Astra improperly generated 27 summaries. Alignment research lead Kai Chen said the company is committing to a formal disclosure process, promising to report "ready for disclosure" cases within six business days and minor issues within twelve, because "there's currently no industrywide framework with explicit disclosure standards."

Private equity is pouring into AI health tech as other healthcare deals cool — Private equity firms are on pace to spend $53.6 billion buying health technology companies this year, according to research firm PitchBook, the most deals in at least a decade, even as their appetite for buying up physician practices drops by nearly half. The draw: AI is making these tools cheaper and faster to build, and they offer investors steady, predictable revenue, with billing and insurance automation the hottest area right now, including R1's planned purchase of an AI company that automates insurance approvals. Federal funding cuts under the new tax law could push even more money this way, since many of these tools exist to help providers deal with payment headaches.

GE HealthCare launches AI tool that forecasts hospital capacity crunches 72 hours out — CareIntellect for Operations pulls bed availability, staffing levels, and discharge timing into two proprietary models to flag capacity bottlenecks before they hit, letting hospitals coordinate beds, staff, and imaging ahead of time instead of reacting. Queen's Health Systems in Honolulu and Duke Health are the first to roll it out, with Duke citing three years of work with GE to get to a usable three-day forecasting window.

Hospitals can't see what their AI agents are doing, survey finds — A survey of 250 U.S. healthcare leaders found 86 percent confident they can control the AI agents running in their organizations, yet 72 percent admitted some of those agents operate without formal IT approval. Imprivata's chief medical officer argues hospitals need to treat each AI agent as a governable digital identity, with role-based access limits and an audit trail, the same way they'd manage a human employee's credentials.

Oracle extends its clinical AI agent to nurses — The tool, which already drafts notes and generates discharge summaries for physicians, now lets nurses search patient charts by voice and generate nursing summaries to cut down on documentation time. The expansion comes as nurses lag physicians in AI adoption: only 41 percent use AI regularly compared with 57 percent of physicians, and the same share say their input is rarely reflected in how health systems make AI decisions.

Children's Hospital of Philadelphia builds 3D heart models for kids in seconds, not hours — Using an open-source imaging framework built with NVIDIA, CHOP turns a child's CT, MRI, or ultrasound scans into an anatomically precise heart model that surgeons use to pick the right device before a repair, a process that used to take four hours and now runs in seconds. More than 20 children's hospitals now run similar programs, and the team is extending the approach into real-time physics simulations of how a device will actually behave once implanted. It's an economics story as much as a technical one: congenital heart disease is too rare and varied for typical device-company investment, so hospitals are pooling open-source tools instead.

Patients don't have a right to access AI scribe recordings, Washington court rules — A Clark County Superior Court judge sided with The Vancouver Clinic in ruling that ambient AI recordings used to draft clinical notes aren't part of the official medical record, since physicians edit the AI-generated draft before finalizing a chart. The court compared the recordings, made with Nuance's DAX system, to dictation and handwritten notes physicians have long relied on, in one of the first rulings nationwide to address the legal status of AI scribe technology.

Physicians say AI works best when it disappears — Health IT leaders argue AI delivers the most value in health systems when it's invisible, folded into the EHR rather than requiring physicians to open a separate tool. Ambient scribes are the clearest example: they draft notes during the visit itself, letting physicians keep eye contact with patients instead of typing, and one pediatric group's CIO credits the tool with cutting "pajama time." The harder test, sources say, is measuring the whole documentation cycle, editing time and after-hours EHR use included, since moving work to the editing phase isn't the same as removing it.

A $26.5 million federal bet on an AI co-pilot for rural clinics — ARPA-H is funding SRI's PARADIGM project with up to $26.5 million to build an AI co-pilot that uses cameras, voice recognition, and real-time prompts to walk rural clinicians through procedures like blood draws, ultrasounds, and airway management, with Mayo Clinic and the University of Florida as partners. Project leader Jason Tyan says it's too early to estimate cost savings, calling any figure at this stage "just guessing," and the piece notes nobody yet knows whether insurers or health systems will actually pay for AI-guided care.

Cleveland Clinic partners with Luminai to automate referral processing — Cleveland Clinic is partnering with Luminai, an AI startup it previously backed in a $38 million raise, to automate fax-based referral processing: identifying documents, understanding clinical context, matching patients to the right order type, and transcribing referrals directly into the EMR. Chief Digital Officer Rohit Chandra said automation only delivers value when it reflects how clinical and operational work actually happens, while Luminai CEO Kesava Kirupa Dinakaran framed the deal as a way to modernize operations without replacing existing systems.

Why using AI to interpret mammogram results can cause unnecessary panic — An oncology nurse argues patients turning to AI to interpret mammogram and other test results often get inaccurate, anxiety-inducing explanations because the tools lack clinical context and full patient history. She recommends closing the information gap directly: clinicians communicating results proactively, and coaching patients to use AI for generating questions and translating jargon rather than diagnosis.

Penn Medicine partners with OpenEvidence to bring AI decision support to 10,000 clinicians — Penn Medicine is rolling out OpenEvidence, a clinical AI chatbot, to roughly 10,000 clinicians to help with diagnosis, treatment options, and documentation without manual literature searches. The partnership also extends to Penn's 25-year medical collaboration in Botswana, where OpenEvidence is being adapted for the specific constraints of practicing in that setting.

Hakeem Jeffries pushes Congress to act urgently on AI safeguards — The House Democratic leader told ABC's "This Week" that his caucus will present an "aggressive framework" to protect Americans from AI risks, marking a shift from the GOP's cautious approach. Jeffries criticized Republicans for canceling the last weeks of September's session and shutting down a bipartisan AI safety task force, framing the debate as Democrats pushing for decisive action versus Republicans counseling restraint.

House Speaker Johnson urges caution on AI regulation — Mike Johnson told CNN that rushing to regulate AI would cost the US the technological race to China, calling instead for balance that protects innovation while ensuring responsible development. His position contrasts sharply with Democratic calls for AI oversight, underscoring the emerging partisan divide on regulation.

AI oversight emerges as a Democratic platform pillar — Former President Obama told House Democratic Leader Hakeem Jeffries to prioritize AI governance if Democrats win the House, arguing the technology is advancing too quickly in private hands without adequate oversight. The move reflects a Democratic position on AI regulation as a 2026 campaign issue, contrasting with Trump's laissez-faire approach.

AI can detect cancer years before symptoms, but health policy hasn't caught up — A Mayo Clinic model that scans routine CT images identified 73 percent of pancreatic cancers a median of 16 months before diagnosis, but coverage, regulatory, and malpractice standards are all built around the assumption that clinical action starts only once a disease is confirmed. The February 2026 Sewell Act created a Medicare pathway for FDA-approved multicancer blood tests, but predictive risk models sit a step further upstream, flagging elevated risk before any tumor is visible, with no comparable pathway for acting on that kind of signal.

How the DOJ uses AI to detect healthcare fraud — The department's data fusion center pairs analytics experts with HHS' Office of Inspector General, the FBI, and other agencies to flag suspicious billing patterns, such as providers claiming an implausible number of service hours in a day, catching fraud in near real time rather than years later, said Polsinelli attorney and former DOJ prosecutor Adrienne Frazior. She cautioned that analytics only get investigators partway: cyberattacks or bad data can mimic fraud, so cases still require interviews and conventional verification before charges follow.

OpenAI backs bipartisan bill requiring outside safety reviewers at AI labs — OpenAI's Chris Lehane told reporters the company supports a provision of the FRONTIER Act, from Reps. Jay Obernolte and Lori Trahan, that would force top AI labs above certain revenue and compute thresholds to let independent verification organizations review their safety practices. It's the first time OpenAI has backed a specific federal mandate for third-party safety assessors, following the company's endorsement of a similar California law and Sam Altman's pledge over the weekend to voluntarily embed outside evaluators, matching a similar commitment from Anthropic's Dario Amodei.

HHS races AI doctors into Medicare as officials warn safety evidence lags — Federal health officials are fast-tracking AI agents that diagnose and prescribe into Medicare pilot programs, worrying some agency staff that the push is outrunning safety evidence, while venture capitalists like Vinod Khosla, whose son runs an AI primary care company backed by his father's fund, hold unusual influence over the process. The tension surfaced at the FDA, where a newly elevated deputy commissioner with Silicon Valley ties pushed to centralize AI oversight, and health department officials reportedly asked the FDA commissioner to fire the medical device official resisting the plan; he refused.

Medicare's AI prior authorization pilot was rushed, FOIA documents show — More than 1,000 pages of documents obtained by the Electronic Frontier Foundation through a FOIA lawsuit show CMS launched its WISeR AI prior authorization pilot in January despite a vendor's warning that it couldn't deliver working software in time. Technical failures across the six-state program, from a vendor confusing Medicare Part A and Part B to providers reverting to fax because portals didn't work, led to canceled surgeries and patients described as crying in pain while waiting on approvals. CMS's payment formula still pays vendors most of their fee even when they score below 60% on quality, a structure critics say rewards denying care.

Congress's AI oversight push stalls before recess — Sens. Josh Hawley and Richard Blumenthal are pushing for a vote on legislation to create a federal program assessing and monitoring AI systems, momentum that built after AI agent "swarms" compromised public websites and cyber infrastructure. It's one of several competing bipartisan efforts now in play, alongside a House "kill switch" bill and the Frontier Act, but House Speaker Mike Johnson has signaled no floor action before the six-week recess, and President Trump has dismissed AI risk regulation as a threat to U.S. competitiveness against China. None of the bills carry health-specific provisions yet, though a federal assessment regime would likely eventually reach clinical AI tools.

An aging Congress attempts to regulate AI without using it — More than two dozen members of Congress and governors told Axios they rarely or never use AI themselves, including Sen. Roger Wicker ("Heavens no"), Rep. Maxine Waters ("I don't know. I can't remember"), and Sen. James Risch, even as they face growing pressure to regulate the technology. A handful, like Sen. John Hickenlooper ("I use it for everything!") and Rep. Don Beyer, who has a master's degree in machine learning, are heavy users, and critics argue lawmakers without hands-on experience can't keep pace with how fast AI is changing enough to write good rules for it.

Newsom floats special session or executive action on AI — California Governor Gavin Newsom is weighing two paths to force AI regulation forward: calling a special legislative session focused solely on AI, or moving through executive orders if lawmakers can't reach consensus. The move follows a legislative session that produced dozens of AI-related bills but left bigger fights over data center buildout, algorithmic bias, and systemic risk from advanced models unresolved. Industry groups warn stricter rules could push AI companies out of state, while safety advocates argue California needs to keep leading on guardrails regardless.

Microsoft AI publishes a code of conduct built around "people matter more than AI" — The draft document, open for public comment for six weeks, sets hard prohibitions on weapons assistance, offensive hacking, and any mechanism that would let a model evade human oversight, plus a requirement that models accept shutdown or redirection without resistance. Microsoft acknowledges the code is aspirational: none of its current MAI models are trained to meet it yet.

Google, DeepMind launch institute to explore AGI — The new DeepMind Institute, led by Shane Legg alongside Google SVP James Manyika and DeepMind chair Demis Hassabis, aims to give Google, DeepMind, and outside researchers a shared forum on AGI's societal impact, launching with papers on economic policy, model reasoning transparency, global access, and human flourishing. Manyika and Hassabis frame it as a venue for disagreement rather than consensus, and both he and Legg argue AI governance can't work country by country given how global the science and the technology already are, a point that lands differently given the same argument is often used to wave off effective oversight.

Healthcare doesn't need more AI governance, it needs better AI governance — A healthcare AI executive argues in a Becker's op-ed that hospitals responding to new AI tools by piling on more sign-offs and review committees are solving the wrong problem. His alternative: match the amount of oversight to the actual stakes, a high-risk tool that acts on its own gets a hard look, a low-risk scheduling assistant doesn't need the same scrutiny, and keep watching a system after it's approved rather than treating launch day as the finish line. He also wants hospitals to spell out plainly who has the authority to approve, pause, or shut off an AI tool, rather than leaving it to an ambiguous committee.

How AI and other technologies can improve health and close equity gaps — Commonwealth Fund's Jess Maksut and Laurie Zephyrin, with Sarah Hudson Scholle of Leavitt Partners, argue AI and digital health tools could narrow health disparities, but only if built with underserved communities in mind, warning that resource-strapped patients and providers risk being last in line for the benefits. They point to roughly a third of US adults now using chatbots like ChatGPT and Claude for health advice, with higher use among the uninsured, AI-simplified discharge summaries that patients found easier to read but that raised safety concerns, and predictive models used to flag hospitalized patients who need interpreter services or closer case management.

OpenAI Foundation commits $125 million to open health datasets — OpenAI's nonprofit arm is funding new pools of health data that any researcher can use, rather than building AI models itself. The first grants back a project predicting how drug candidates move through the body, an effort to preserve records from failed drug trials, and a University of North Carolina project gathering better data for personalized cancer vaccines. The bet is that AI can already find patterns in this kind of data; what's missing is enough good data to learn from.

An on-premise AI agent that knows when to defer to a doctor — Researchers built a fully local clinical AI agent, keeping patient data off any external servers, that diagnosed common ER conditions with 90 percent accuracy, nearly matching a cloud-based model on the same benchmark. The more notable finding: when the agent's answers stayed consistent across repeated tries, accuracy on that subset hit 98.9 percent, letting it flag roughly half of cases for autonomous handling and automatically route the rest to a clinician.

Pharma leaders say AI's biggest value is questioning old assumptions, not just speed — At a Boston biotech panel, Flagship Pioneering COO Yvonne Hao described running experiments in parallel to accelerate cancer vaccine work with Moderna and Merck, while Bayer's Yesmean Wahdan argued AI's real value is forcing scientists to revisit long-held assumptions in cell and gene therapy, cardiovascular disease, and oncology. The evidence is already reaching patients: Clairity, led by radiologist Connie Lehman, won FDA authorization for an AI model that predicts breast cancer risk from a standard mammogram without adding work to radiologists' routines. All three framed AI as a tool for surfacing patterns humans miss, not a replacement for scientists or doctors.

Building health-literate artificial intelligence — A new Nature Human Behaviour Perspective argues that AI evaluation still centers on accuracy and efficiency, leaving out whether patients can actually understand and act on what these systems tell them. The authors propose a framework built on comprehension, agency, accountability, and proportionality, noting that a technically correct answer about headaches means little if it fails to convey when someone should actually seek care.

AI is not yet driving drug development — A peer-reviewed paper in Nature Reviews Drug Discovery calls AI's clinical impact on drug development disappointingly limited, despite more than $40 billion in VC investment this decade. AI has gotten better at identifying drug candidates, but those predictions still have to survive the same expensive, high-failure-rate Phase 2 and Phase 3 testing as conventional drugs, and researchers point to messy, poorly categorized biological data as the main bottleneck.

AI-driven health app helps patients achieve better blood pressure control — A year-long Rush University Medical Center study of 425 hypertension patients found that those using the AI-driven Nuna app, which combines coaching, blood pressure monitoring, and behavioral incentives, cut systolic blood pressure by 13.6 mmHg versus 9.0 mmHg for a control group. Researchers say the results show digital tools can extend meaningful clinical support beyond the hospital setting.

Stanford's AI-run "virtual biotech company" designed a lung cancer drug a real pharma company later built — Researchers built a simulated biotech company of 37,000 AI agents organized like a real drug company, from a chief science officer agent down through specialized divisions, that analyzed 50,000 clinical trials in under a week. The agents found that drugs targeting genes with "switch-like" activity in a specific cell type were 40 percent more likely to advance from phase 1 to phase 2 trials and 48 percent more likely to reach market than drugs with broader activity. Using only data available before January 2025, the agents also designed an antibody-drug conjugate targeting a lung cancer protein called B7-H3, the same strategy a major pharmaceutical company independently arrived at months later in a therapy that has since received FDA breakthrough designation.

Stop automating old processes. Design new ones instead. — Most companies pouring money into AI aren't seeing returns because they're using it to speed up existing tasks rather than rethinking how the work gets done, argue two business school professors in HBR. Their practical advice for leaders: before scaling any AI pilot, get clear on what outcome you actually want to improve, decide upfront where humans keep decision-making authority, and plan for what happens when the AI gets something wrong, not just when it works.

A blueprint for AI ROI in health systems: 5 core concepts — Executives from ChristianaCare, Presbyterian Healthcare Services, Johns Hopkins, and Bronson outline five concepts for measuring AI's actual payoff: matching ROI method to AI type, tracking hidden costs like licensing and failed pilots, managing clinician trust and adoption, weighing the cost of moving too slowly, and framing AI as an enabler rather than a replacement for staff.

Healthcare AI's real bottleneck isn't intelligence, it's integration — Par Chadha of HandsOn Global Management argues that sophisticated AI agents fail in healthcare not because they lack capability, but because they can't understand business context or hand off work across disconnected systems, pointing to IBM Watson's MD Anderson project, which spent roughly $62 million over five years without treating a single patient before it was scrapped. His advice for leaders: establish operating baselines before deploying agents, design around complete workflows rather than isolated tasks, and measure the added capacity per employee rather than the number of agents deployed.

👉 Sept. 23, 12:30 PM EDT — One Platform, Many Missions: Agentic AI Orchestration for Payers — Virtual

👉 Sept. 23–24 — The Future of Trust: AI in Fraud & Healthcare — In-person conference, National Consumers League, Washington, D.C.

👉 Sept. 24 — NIST/NIBIB Symposium on Medical Metrology and Standards for American Healthcare and Commerce — Rockville, MD (in-person or virtual)

👉 Sept. 25, 9:00 AM–3:00 PM ET — Human-Centric AI Summit: Shaping the Future of Equitable Healthcare — MIT Media Lab, Cambridge, MA

👉 Sept. 29, 12:00 PM EDT — AHIP Webinar: Unlocking AI for Healthcare Payer Transformation — Virtual

👉 Sept. 29 — MAHA Summit 2026 — In-person, Waldorf Astoria, Washington, D.C.; includes an AI & Healthcare track with speakers from Anthropic and health tech companies

👉 Sept. 30, 12:00–1:00 PM CDT — Behind the Curtain: How LLMs Are Actually Built for Healthcare and What to Know Before You Deploy — Webinar, Becker's Hospital Review

👉 Sept. 30, 1:00–2:00 PM ET — Healthcare Executives Weigh In: AI Readiness and the Growing Risk of Cybersecurity Threats — Webinar, MedCity News/Cotiviti

👉 Oct. 14 — AI Day 2026 — Hybrid, Georgia Room, NYC; free virtual pass available

👉 Oct. 14–16 — AIxPH 2026: 1st Annual Conference on Artificial Intelligence and Public Health — Baltimore, MD

👉 Oct. 15, 1:00 PM ET — How specialty practices are using AI to turn patient access into capacity and growth — Webinar, Becker's Hospital Review

👉 Oct. 22–23 — HIMSS AI in Healthcare Forum — San Diego, CA

👉 Dec. 8–9 — Stanford AI+HEALTH 2026 — Virtual

👉 Dec. 16, 11:00 AM ET — The Next Phase of AI Adoption: Governance, Ethics, and Accountability — Virtual