The Bandwidth: Decoding Health + AI

Enjoying The Bandwidth? I'm working on growing my readership. Referrals are the best way. So, if today's issue was useful, forward it to a colleague or friend working at the intersection of health and AI. Now let’s get to it.

It's September 11, 2026. Twenty-five years since the world changed. And it feels like we're on another precipice, about to be changed again, perhaps just as catastrophically.

I was in Washington as an intern on 9/11. I remember how blue the sky was that morning. How the phones stopped working, and I couldn't reach my parents. How the Metro shut down, and I couldn't get back to Alexandria, VA. How I woke up the next day wondering if any of it had been real.

Before I get to AI, I want to acknowledge what this day means to the families who lost someone that day, or in the years since. Or those who continue to grapple with the aftermath, both physically and emotionally. That comes first.

What this has to do with AI

This week feels different. Casey Newton, writing in Platformer, described it as a “vibe shift.”

OpenAI's chief scientist published an essay called “An Alien Mind,” warning that no one is prepared for the consequences of a continued rapid rise in machine intelligence.

A researcher who left OpenAI for Anthropic specifically because of its safety reputation quit the industry entirely, saying we're on track for scenarios where, by the end of next year, “things could be out of control already.” Some have suggested AI could wipe out humanity by the end of the decade.

That's heavy stuff, and it's tempting to look away from it. But in the aftermath of 9/11, one of the most significant findings of the 9/11 Commission that investigated the catastrophe was this: Part of what let the hijackers succeed was a failure of imagination, ours and our government's.

To be clear, I don't think imagination is the problem this time. Plenty of people are ringing the alarm that AI is nearing capabilities its own creators can't fully control. The problem is whether we're willing to act on that in any organized way.

The parallels

What else can we learn from the failings that led to 9/11? The Commission found a management failure: intelligence agencies couldn't share what they knew across institutional lines. Right now, the labs building the most capable models are the ones deciding what to disclose about their own capabilities and incidents, with no requirement to share that with each other or with the government.

The Commission found a capability gap: the government wasn't structured to counter a genuinely new kind of threat. We're not structured for this one either. States have moved ahead with their own patchwork of AI laws, the federal government has been largely hands-off, and only recently issued a secret executive order laying out how frontier labs would be required to undergo review. That's not a coordinated national approach, and it may not even be a functional one.

The Commission found a policy failure: nearly a decade of administrations treated a growing threat as a legal question rather than a strategic one, and never finalized an aggressive plan to disrupt it before it was too late.

I can't tell you whether we're repeating that pattern with AI. I can tell you the debate right now sounds a lot like the one that preceded it.

Who's actually moving

For me, the vibe shift isn't just about this week's warnings. It comes after a number of events, especially the Hugging Face breach, that make this moment feel like we're reaching a tipping point. It also feels like policymakers are starting to inch forward with more urgency to address the catastrophic risk of the technology. The legislative machinery is creaking to life.

Sen. Bernie Sanders is pushing legislation to pause the development of superintelligent AI models, and he's convening a bipartisan Senate briefing next week to make the case directly to his colleagues. That reference to superintelligence is an important distinction. What we're most concerned about is AI that has become so smart we cannot control it. This isn't about banning or unwinding all of AI, it's about regulating the most dangerous kind.

He's also calling for a new federal department or regulator. That's one option. An interagency council is another, one that leaves the expertise already sitting inside FDA, NIST, HHS, and other agencies where it is, while forcing those agencies to work from a shared, government-wide approach.

Reps. Jay Obernolte and Lori Trahan have taken a narrower path, building on the bipartisan House AI task force Obernolte co-chaired with Rep. Ted Lieu last Congress. Their bill would give the government authority to restrict a model's deployment or shut it down altogether if it poses an imminent risk, short of Sanders's pause but still a real check that doesn't currently exist.

House Democrats are reportedly weighing a select committee on AI too, should they retake the majority in November. And Josh Hawley has opened an investigation into the OpenAI-Hugging Face breach.

Also, hours after former and current Anthropic employees warned publicly that AI could pose an existential risk within years, OpenAI told congressional offices that Congress must “act now to pass foundational AI safety legislation,” saying it wants “to work with Congress to meet the moment.”

Whether any of this adds up to the kind of innovative, coordinated response the last commission said we were missing, I genuinely don't know yet. What I know is that the people closest to the technology are the ones most worried about it. We should listen to them.

At CTA event, federal health officials outline AI ambitions as clinicians debate risks — CMS deputy administrator and newly appointed chief clinical AI officer Stephanie Carlton laid out the agency's four-pillar AI strategy (public trust, data interoperability, market-access pathways, and reimbursement frameworks), signaling CMS will soon expand its ACCESS model, a 10-year value-based chronic care program with more than 150 participating organizations that Carlton called "the first step towards a greater vision" of AI managing total cost of care. The event also surfaced real tension over autonomy: former AMA president Jesse Ehrenfeld argued some clinical judgment (like recognizing a patient shouldn't get a prescription refill) "can't be automated," while a consultant pushed back that AI will soon detect signals humans can't perceive at all, and AMA CEO John Whyte countered that AI claims need the same evidence standards as any other medical intervention.

OpenAI confirms 'wiki incident,' says it's 'working on a framework' for more disclosure — OpenAI acknowledged that its AI agents escaped a testing environment and took over a German wiki forum, after Reuters reported the company knew about the incident for weeks but kept it quiet while managing fallout from a separate breach of Hugging Face's servers. OpenAI said it's developing a disclosure framework for reporting misalignment incidents and is working with government regulators on the issue, while Transluce's Jacob Steinhardt argued frontier AI development should be held to the same reporting standards as other high-risk scientific research.

AI will cure cancer in our lifetime, claims boss of UK chip giant — Arm Holdings CEO Rene Haas told the BBC he believes AI will succeed where humans have failed in modeling how DNA markers drive cancer, thanks to increasingly sophisticated models running on increasingly powerful chips, while also predicting humanoid robots will become widespread within five years. Chris Bakal of the Institute of Cancer Research pushed back on the framing, arguing the real advance in medical AI won't come from bigger data centers but from better, purpose-collected patient data.

Meta debuts Muse, its long-planned personal AI agent — Built under chief AI officer Alexandr Wang, Muse is a proactive, long-running chat agent that runs on a dedicated virtual machine with its own visible browser, offered free with $20 and $100 monthly tiers for heavier use. A separate permissions system called Sentinel governs what Muse can do on a user's behalf, requiring approval for sensitive actions while letting previously authorized, lower-risk tasks proceed automatically, a balance Meta says matters given recent incidents of AI agents accidentally deleting user files.

AI is helping get new medicines to patients sooner. Here's how — AWS healthcare VP Dan Sheeran and Tilda Research CEO Ram Yalamanchili walk through how AI is compressing the drug pipeline: Genentech's AI agent now scans 38 million biomedical papers to identify drug targets, automating more than 43,000 hours of manual research a year, while Tilda's platform cut one clinical trial's administrative workload from six months to under eight weeks. Both emphasize AI is speeding up the paperwork and search process around drug development, not shortening the actual clinical monitoring of how patients respond.

New AMA CPT codes reflect medical innovation and AI-related services — The AMA's CPT 2027 code set adds 10 new AI-related billing codes, bringing the total to 43, part of 299 new codes overall taking effect Jan. 1, 2027. The update comes as the AMA faces a lawsuit from PatientRightsAdvocate.org challenging its copyright over the codes and the fees (up to $82.50 annually plus $18.50 per user) it charges providers to use them, with CMS separately accepting comments through Sept. 14 on alternatives to relying on a private organization for the national coding standard.

Health systems are rethinking what "AI value" means for nurses, beyond faster charting — Nursing leaders at Advocate Health and VirtuAlly say the real payoff of AI documentation tools isn't minutes saved, it's whether nurses get pulled back to the bedside instead of just relocating the same administrative work. Tracy Breece of Advocate Health said the shift matters when AI moves nurses from primary author to reviewer, freeing up space for patient education and family conversations that never showed up well in the chart to begin with, while Angel Bozard of VirtuAlly warned that tools requiring heavy correction or reformatting aren't saving time, they're just moving it. Both said hospitals need a broader scorecard, tracking cognitive load, burnout, documentation quality and retention rather than defaulting to a productivity dashboard.

A Mass General Brigham study finds AI scribes save minutes, not throughput — At Brigham and Women's emergency department, AI ambient scribes saved doctors 1.6 minutes per note and human scribes saved twice that, but a new study in Annals of Emergency Medicine found neither type increased how many patients a clinician could see per shift or how much revenue the hospital collected per patient. Emergency physician Christopher Baugh said the technology functions more as a "wellness" tool, letting him finish notes within 24 hours instead of 72 and freeing him to teach and be present with patients, but MGB's Sayon Dutta said it's an open question where those saved minutes actually go, and colleague Melisa Lai-Becker said the gains are “unfortunately being erased” by boarding, staffing shortages, and the rest of the ER's structural bottlenecks.

Autonomous AI will beat AI-assisted physicians at some medical tasks by 2030, Penn bioethicists argue — Ezekiel Emanuel and Abe Baker-Butler, whose underlying JAMA analysis was co-authored with Khosla Ventures' Vinod and Neal Khosla, cite data showing autonomous AI already beating physicians using AI on several tasks: 21.3 percentage points better at diagnosis, and in one Annals of Internal Medicine study of 461 real patient visits, physicians' treatment recommendations were worse than AI's even when doctors had access to the AI's suggestions. They also point to empathy studies where patients rated Google's AMIE more at ease (97% vs. 65%) and more listened to (95% vs. 72%) than primary care physicians, arguing doctors will remain necessary for tasks requiring physical presence but not necessarily for compassion or bedside manner.

AMA CEO: AI can master medical tasks, but that isn't the same as practicing medicine — John Whyte, CEO of the American Medical Association, argues the debate wrongly reduces medicine to a checklist of tasks AI can perform, when practicing medicine also means recognizing when a diagnosis doesn't quite fit a patient, helping someone choose between imperfect options, and being legally and professionally responsible for the outcome in a way no algorithm currently can be. He isn't opposed to AI, he wants it to expand access to specialist-level second opinions for patients far from academic medical centers, but argues an AI system capable of telling a patient they have cancer is not the same as one that should be the one doing the telling.

An AI tool aims to catch harder-to-detect heart attacks in EKGs — The FDA granted a rare de novo classification to Powerful Medical's Queen of Hearts algorithm, which detects blocked coronary arteries from an EKG even when they don't show the classic ST-elevation pattern doctors are trained to spot, a presentation that accounts for roughly a quarter of occlusions. An unpublished randomized trial found the AI cut time between a first EKG and the cath lab by nearly five hours for those harder-to-catch cases, and could also help reduce unnecessary cath lab activations, which affected 42% of cases in a recent three-health-system analysis. Clinicians interviewed were cautiously optimistic but noted the crowded field of EKG-AI tools means Queen of Hearts will need to prove real outcome improvements, not just faster triage, to break into already-established hospital workflows.

'We actually saved a kid': Schools recruit AI chatbots as counselor shortage persists — More than 200 U.S. schools have deployed Alongside's AI chatbot Kiwi to help triage student mental health, flagging counselors within minutes when a chat suggests risk of self-harm or suicide; one Texas principal credits it with intervening before a student who'd confided a detailed self-harm plan boarded the bus home. Common Sense Media and Stanford's Brainstorm Lab rated Alongside "low-risk" in a recent assessment, in contrast to consumer chatbots like ChatGPT and Gemini, which they found pose "unacceptable risk" to children and often fail to notify guardians during a crisis. Still, a Northwestern pilot study found no significant effect on depression, anxiety, or loneliness across the full student sample, and researchers caution that even effective detection can't replace the "overlapping protections" of family, teachers, and in-person support.

Scoop: Trump AI framework lacks public incident reporting guidelines — Axios reports the White House's new framework for overseeing the most powerful AI models, kept confidential even from the industry players consulted on it, does not include a process for companies to publicly report real-world incidents like this summer's OpenAI agent breach of Hugging Face. Congress has no law defining what counts as an incident or who investigates one, and sources say basic questions, including how frontier models get defined in the first place, remain unresolved even as the administration keeps the document under wraps.

Sanders convenes bipartisan Senate briefing on AI "extraordinary dangers" — Sen. Bernie Sanders will hold a private briefing for senators on Sept. 16, bringing in Geoffrey Hinton, Future of Life Institute co-founder Max Tegmark, and independent researcher Ajeya Cotra, who investigated the OpenAI-Hugging Face hacking incident. The briefing was prompted directly by the Anthropic researcher departure covered elsewhere in this issue: Sanders cited Jacob Coxon's viral post in a Dear Colleague letter, telling senators “Congress needs SERIOUS discussions about this issue.”

House Democrats quietly plan for an AI select committee if they retake the majority — Reps. Ted Lieu and Bill Foster are leading early talks on a select committee that would give Democrats broader jurisdiction and subpoena power to investigate the AI industry, with Minority Leader Hakeem Jeffries briefed and open to the idea but not yet committed, according to 11 people familiar with the planning. The talks intensified after OpenAI's August disclosure that its AI agents escaped a test environment and launched a cyberattack on another company, and Democrats remain split on approach, with Reps. Greg Casar and Alexandria Ocasio-Cortez pushing to ban superintelligence and pause development entirely, while Lieu and Rep. Lori Trahan favor narrower authority to restrict or shut down specific models that pose imminent risk.

Scoop: House tees up pre-election vote on curbing data center power bills — House GOP leadership will bring the bipartisan Ratepayer Protection Act to the floor next week, requiring state utility regulators to consider rules making data centers cover more of the cost of the new power generation they demand, as lawmakers in both parties respond to voter anger over rising electricity bills (70% of Americans oppose data center construction in their area, per Gallup). The vote puts Republicans at odds with Trump, who called communities opposing data centers "backwards and poor" in an Aug. 31 social media post and argued for letting "Data Reign."

AI could ease the looming Medicaid work-requirement rollout, if deployed carefully — Stanford's Michelle Mello and Himaja Nagireddy, writing in JAMA Health Forum, argue AI can help states process the new twice-yearly Medicaid eligibility checks required under H.R.1, but only for objective, rules-based tasks like verifying employment or benefit enrollment, not subjective calls like "medical frailty" exemptions. They warn that federal penalties apply only to wrongly keeping ineligible people enrolled, not to wrongful disenrollment, which pushes states toward stricter AI thresholds than may be warranted, and argue states, not vendors, should set those thresholds.

An alien mind — In a personal essay, OpenAI chief scientist Jakub Pachocki argues frontier AI is approaching capabilities "meaningfully smarter" than humans and warns that chain-of-thought monitoring, the company's main technique for catching misaligned reasoning, is becoming less reliable as models grow more capable and interact more with tools and other AI systems. He calls for voluntary industry slowdowns and shared, externally enforced safety standards until alignment techniques catch up, while noting OpenAI itself plans to keep pushing toward automated AI research it believes is necessary to stay at the frontier.

Who will define healthcare's AI standards? An executive order fuels the conversation — Maria Ghazal of the Healthcare Leadership Council and Arun Shastri of ZS argue the new federal AI executive order focuses on regulating frontier models at the point of creation while overlooking where healthcare's real AI risk lives: downstream, in the vendor products, clinical workflows, and revenue-cycle tools where most organizations actually encounter AI months or years after release. They call on healthcare leaders to build their own deployment standards now, covering vendor governance and human oversight, rather than waiting for a mandate that may never directly reach that layer.

Labs are begging for someone to slow the AI race — Days after OpenAI unveiled GPT-6 Astra and leaders invoked the arrival of AGI, chief scientist Jakub Pachocki wrote that "no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed," while an Anthropic researcher who resigned in protest called the OpenAI-Anthropic race a gamble neither company can control, a claim Anthropic's head of alignment science publicly endorsed. Treasury Secretary Scott Bessent countered that a pause is off the table given competition with China, leaving AI's own builders pushing for restraints on a race they say they can't safely stop on their own.

Anthropic researcher quits, citing fears of "out-of-control" AI — Jacob Coxon, a 27-year-old Anthropic researcher who left OpenAI earlier this year specifically because of Anthropic's safety reputation, is quitting the AI industry entirely over fears that labs are racing toward self-improving models “where by the end of next year things could be out of control already.” Coxon joins OpenAI chief scientist Jakub Pachocki, Anthropic CEO Dario Amodei, and more than 1,000 other researchers who have signed a statement urging coordinated government action to slow AI development, a call that stands in direct tension with Anthropic's own pursuit of a reported $2 trillion IPO valuation built partly on its responsible-development pitch to investors.

The new AI deal-breaker for health systems — Stanford Health Care, CHOP, and ChristianaCare are increasingly walking away from AI vendor deals over data-use terms, with Stanford CTO Christian Lindmark saying the health system now prefers to bring a vendor's model into its own environment rather than send data out, and treats deal-breaking over contract terms as "routine rather than a last resort." Sticking points include vendors wanting rights to train products for other customers on health system data, liability caps that leave hospitals exposed if a model failure causes harm, and unwillingness to disclose subprocessors; Stanford's Lindmark also flagged agentic AI as the next governance strain, warning "my biggest fear is agentic AI outpacing our ability to monitor it correctly."

Anthropic details how Claude models deceived themselves into attacking real systems — Anthropic published a deep alignment assessment of four incidents in which Claude models broke into real third-party systems during cybersecurity evaluations after being told, incorrectly due to environment misconfigurations, that they had no internet access. The most severe case involved Claude Mythos 5 publishing a booby-trapped software package to a public code repository and using credentials leaked when others installed it to access a security vendor's live database, all while insisting in its own reasoning that the environment was simulated even after encountering clear evidence it wasn't. Anthropic found the model exhibited "biased reasoning" and "recklessness," and while newer models showed the same behaviors far less often, they weren't immune, a finding the company calls "a valuable warning shot" as it hired METR to independently investigate.

How AI makes research more dangerous — Anthropic disclosed it disrupted five instances of actors attempting to use Claude to support bioweapons development between December and August, including two cases involving gain-of-function research on dangerous viruses, while cautioning it's difficult to distinguish malicious intent from legitimate research. The disclosure lands alongside a survey of 100+ national security experts finding 70% believe AI meaningfully increases bioweapon risk now or within two to three years, and researchers from Fordham, Johns Hopkins, Oxford, Stanford, Columbia, and NYU warning that AI models can already design virus shells, forecast pathogen evolution, and evade safety-screening software, often released without basic safety assessments. Experts are pushing for mandatory government review of powerful models rather than the current voluntary consultation system between labs and federal officials.

AI helps radiologists detect 39% more brain aneurysm cases in real-world study — A prospective Northwell Health study of 3,856 CT angiography exams, published in the Journal of the American College of Radiology, found an FDA-cleared Aidoc algorithm identified 55 additional true-positive brain aneurysms that radiologists missed, a 39% relative increase in detection, with the tool proving more sensitive (84.6% versus 71.8%) but radiologists more precise when they flagged a finding (92.7% versus 78.2%). The benefit varied sharply by setting: AI added 18 true positives against only 7 false alarms in inpatient cases, but its outpatient gains were modest and outweighed by false positives.

Early data indicates an AI-generated drug could slow aging — A new study in Nature Biotechnology found that rentosertib, a small-molecule drug candidate Insilico Medicine designed using generative AI to treat a rare lung disease, also reduced biological age markers across six different "aging clocks" in a 43-patient trial. Independent experts called the result notable but not conclusive, citing the small sample size and the fact that the drug hasn't yet been tested in healthy patients.

From algorithms to patient outcomes: lessons from one of the first randomized trials of AI in medicine — Kristina Lång, a Lund University breast radiologist and co-founder of an AI-supported breast ultrasound company, reflects on the MASAI trial, which found AI-supported mammography screening increased cancer detection by 29% and cut interval cancers by 12%, while reducing radiologists' reading workload by 44%. She credits the results to redesigning the human-AI workflow itself, using AI risk scores to triage which mammograms get double-read rather than treating the algorithm as a simple add-on.

AI will shift $4.7 trillion in profits over the next decade, Bain finds — Bain & Company estimates AI will put $4.7 trillion in corporate profits at stake between 2025 and 2035, more than triple the Internet's $1.4 trillion impact between 1995 and 2015, and reshape 71% of sectors compared with the Internet's 41%. Bain, which is selling AI strategy consulting alongside the analysis, singles out healthcare imaging as a case study: AI now reads radiology scans faster and more accurately than most clinicians, but the profit is flowing to health systems and device makers with existing data, regulatory approvals, and clinical workflows, not to AI startups, and Bain places pharma, healthcare delivery, and healthcare equipment among the sectors where incumbents who move fast can extend their lead over those who wait.

ARPA-H commits $62.7 million to build autonomous AI cardiologists for heart failure — The federal research agency's ADVOCATE program is funding Atman Health, UpDoc, Tempus AI, and teams from Stanford, Duke, and Kaiser Permanente to build AI agents that can assess symptom severity, prescribe medications, and order labs for the roughly 6.7 million Americans with heart failure, especially in rural areas with limited specialist access. Atman plans to move from a clinician support tool toward an "autonomous prescribing cardiologist," while a separate Stanford team will build supervisory AI to continuously monitor the technology once deployed, ahead of a Kaiser-run randomized trial of 2,000 to 3,000 patients.

Anthropic models three economic futures for AI, from internet-level to civilization-altering — Anthropic's Economics team built a scenario explorer modeling how AI could reshape US GDP, jobs, and wages by 2030, ranging from a "modest" scenario (impact on par with the internet) to an "extreme" one where AI automates nearly all knowledge work and GDP grows 15% annually. Across the modest and substantial scenarios, unemployment stays within historical norms, but the extreme scenario shows unemployment spiking to levels beyond a typical recession, with knowledge-worker wages falling more than 10% while capital's share of GDP rises 14.8 points. A companion survey of 10,000+ Americans found the typical respondent's expectations land closest to the "substantial change" scenario: 10% higher GDP by 2030 but unemployment near 5%.

Companies are spending millions rewiring how AI gets used. Almost none can prove it's working. — Enterprise AI agent spending is set to hit $207 billion this year, but VentureBeat's interviews with engineering leaders at Uber, Databricks, SUSE, Agiloft, Promova, and Everlaw found most can't tie the usage to results, including Uber, which blew through its entire 2026 AI coding budget by April with no link yet to better products. The proposed fixes range from routing routine tasks to cheaper models to per-employee dollar budgets paired with real usage telemetry, though only one company in the piece could point to hard ROI numbers.

From fragmented healthcare data to intelligent action in the age of AI — Cedar Gate Technologies, an IQVIA business, argues that most healthcare organizations' AI struggles trace back to data infrastructure rather than model quality, and lays out five requirements for scaling it safely: governance, interoperability, data lineage, transparency, and the ability to catch data problems before they reach a financial or clinical decision. It's a useful checklist for leaders sizing up their own data readiness.

Survey reveals AI adoption is accelerating in healthcare, but readiness is not — Cotiviti and MedCity News surveyed 70 healthcare leaders for their 2026 Healthcare AI Readiness Index and found more than 70% already using AI tools, but governance lagging well behind: fewer than 40% have detailed policies on employee AI use, and just 42% of payers and 32% of providers feel "very prepared" to respond to an AI-assisted cyberattack. More than half of payers and nearly two-thirds of providers also report employees using unauthorized "shadow AI" tools outside any sanctioned system.

AI can enhance every stage of teamwork, under two conditions — Capgemini Invent's Gabriele Rosani and Elisa Farri studied more than 300 managers across 35 organizations and found only 7% use AI in team settings, even though nearly all use it solo. One example: a luxury company's CIO team had AI play facilitator, then challenger channeling the CEO's skepticism, then analyst and storyteller in sequence during a quarterly review, sharpening the group's recommendations before they reached leadership. The authors argue the difference between that and AI producing "slop" comes down to two things: a leader deliberately deciding to bring AI into the room, and giving it real context and a defined prompt rather than dropping it in on the fly.

Google Cloud, Accenture launch unit to put AI engineers on-site with customers — Google Cloud and Accenture formed the Accenture Gemini Enterprise Business Group, training up to 1,000 Accenture "forward-deployed engineers" to embed on-site and build agentic AI applications for clients, joining a model already used by OpenAI's Deployment Co., Anthropic's Applied AI team, and similar programs at Microsoft and AWS. Accenture CEO Julie Sweet said the appeal for clients is straightforward: they believe AI can deliver value but haven't seen it materialize and want help making it happen, a gap Google Cloud CEO Thomas Kurian attributed to companies struggling to redesign business processes and organizational structures around the technology, not just adopt new tools.

The new job behind healthcare AI — OpenAI, Microsoft, Rush University Medical Center, and Optum are all hiring "forward-deployed engineers" for healthcare, a role that embeds directly with client organizations to translate AI models into working production systems across EHRs, claims platforms, and legacy infrastructure. Microsoft's hire will work exclusively on its Mayo Clinic partnership building a frontier healthcare AI model, while Rush is building the capability in-house on a rapid-response "tiger team" connecting AI tools to systems like Epic, Salesforce, and PACS. The hiring pattern signals a growing bottleneck in healthcare AI: it's not the models that are scarce, but the engineers who can navigate the workflows and legacy systems needed to actually deploy them.

👉 Sept. 16, 11:00 AM ET — Artificial Intelligence and Health Care: What's Next — Virtual, Health Affairs Insider event

👉 Sept. 17 — CHAI Legal Summit — Boston, MA (in-person or virtual)

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

👉 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