The Bandwidth - Weekly AI Roundup

If anyone still doubts that AI is already reshaping healthcare, three stories from this week make the trend impossible to ignore.

1️⃣ Sam Altman is now the most influential person in U.S. healthcare, according to Modern Healthcare. Not a biotech CEO. Not a major health system leader. The head of OpenAI. Why Altman? Because ChatGPT has become integral to the strategic plans of hospitals, insurers, and digital health companies. And because consumers have made it one of the most widely used health tools in the country. Altman noted that health advice is one of ChatGPT’s most popular use cases, and OpenAI explicitly prioritized improving that function in GPT-5. Side note: ChatGPT was the most-downloaded free iPhone app in the U.S. in 2025, beating Threads, Google, TikTok, and WhatsApp, underscoring how often people now reach for a chatbot before a browser. When the country’s most frequently used health information interface is an AI model, the person shaping one of the dominant models becomes a healthcare leader by definition.

2️⃣ Microsoft’s analysis of 37 million Copilot conversations also confirms this pattern: health is the top use case. Microsoft AI researchers examined how people used Copilot across time, context, and device. The strongest signal came from mobile.
On mobile, users consistently asked more health and wellness questions than anything else — more than coding, travel, or general search — and that pattern was stable across the year. Desktop behavior varied widely, but mobile never did. The takeaway: AI engagement isn’t static; it depends heavily on context. On desktop, AI acts like a colleague. On mobile, it becomes a confidant — and increasingly, the place people turn first for guidance on health and wellbeing.

3️⃣ Meanwhile, Microsoft released GigaTIME, a model that extracts advanced tumor insights from a ten-dollar pathology slide. GigaTIME was trained on 40 million cell samples and benchmarked across 24 cancer types, surfacing over 1,200 immune patterns and creating a virtual tumor population of 300,000 cases. This signals that AI is beginning to push the clinical frontier — faster and at dramatically lower cost than traditional methods. As the research team put it: by learning to translate across biological modalities, GigaTIME is an early step toward “learning the language of patients,” with the long-term goal of enabling “virtual patients” capable of forecasting disease progression and treatment response. The model shows how multimodal AI could scale real-world evidence generation in ways clinicians and researchers have never had access to before.

So what to make of all this? Taken together, these stories point to a clear reality: AI is already the front door for patients and clinicians, and the accelerator for biomedical research.

OpenAI releases GPT-5.2 (“garlic”) and announces a $1B partnership with Disney.
GPT-5.2 brings improved agentic reasoning and enterprise-tier pricing. Disney becomes Sora’s first major content partner, deploys OpenAI tools company-wide, and invests $1B for a strategic stake. This marks a shift from enterprise experimentation to full-scale adoption.

OpenAI releases first “State of Enterprise AI 2025” report, showing rapid real-world adoption and productivity gains.
The report finds enterprise AI usage exploding — weekly ChatGPT Enterprise messages are up roughly 8× and roughly 75% of surveyed workers reporting that AI improved the speed or quality of their work and saving 40–60 minutes per active day on average. It also highlights widening gaps between “frontier” users and median adopters as companies integrate AI deeper into workflows and bespoke Custom GPTs grow ~19×, underscoring that organizational readiness, not model performance, is the primary constraint. Though let’s be honest, model performance is still a rate-limiting factor.

Halle Tecco’s 2025 health tech recap: AI moved from pilot to production.
Digital adoption held firm even in tough conditions, and AI-native companies captured outsized value. Health tech is splitting into “haves” and “have-nots” heading into 2026.

Meta’s “Avocado” AI strategy falters.
Internal teams are reportedly frustrated with missed targets and repeated resets. This instability could make health organizations cautious about over-relying on Meta’s AI stack.

Accenture partners with Anthropic to deploy Claude across enterprise consulting.
Claude will now sit in workflow redesign, client delivery, and industry-specific deployments. For health systems, this means Claude becomes a default part of consulting-led AI modernization.

President Trump issues an executive order to override state AI regulations.
The order directs DOJ to challenge state AI laws and empowers the Department of Commerce to restrict broadband funds to states with aggressive AI rules. It aims to centralize national AI policy by weakening state-level protections. This could undercut state consumer and patient-safety rules, creating uncertainty for health systems deploying clinical or patient-facing AI.

Forty-two state attorneys general demand new safety controls for AI chatbots.
Separately, 42 state AGs warn that “sycophantic and delusional outputs,” especially when aimed at children, may violate product safety, consumer protection, and online-privacy laws. In a sign that states are not ceding any ground in overseeing AI, they ordered major AI companies to implement safety tests, warning labels, independent audits, and incident reporting by Jan. 16, 2026.

House Democrats launch a Commission on AI and the Innovation Economy.
The commission will shape federal AI policy, workforce strategies, and economic competitiveness. Health care should expect more structured oversight of algorithms touching patients or public programs.

JAMA Forum: How AI could solve medicine’s productivity problem
Scott Gottlieb argues AI can break healthcare’s productivity trap, only if FDA modernizes oversight and CMS fixes reimbursement so efficiency gains aren’t penalized or discouraged.

Civil rights groups warn that AI used in utilities and benefits could harm Black and Latino households.
Advocates say automated decision-making in water, electricity, and public benefits risks replicating discrimination at scale. This scrutiny will inevitably extend to Medicaid eligibility systems and care-management algorithms.

NAACP calls for an equity-first approach to AI governance in healthcare.
The organization urges mandatory bias testing, transparency requirements, community oversight, and stronger accountability mechanisms for AI used in health settings.

AI for Everyone
Andrew Ng’s free introductory course (~Coursera) explains AI in plain English and is designed for people who feel overwhelmed or unsure where to start. It’s one of the cleanest on-ramps for building team-wide AI literacy without coding or heavy math.

The Batch
If you’re not subscribed to The Batch, it’s one of the most approachable ways to stay current on AI without drowning in technical jargon. Each issue summarizes the major developments of the week in clear, practical language — ideal for leaders who want steady AI literacy without the overwhelm.

How to Future-Proof Your Job: The Four Skills AI Can’t Replace
This piece breaks down the human capabilities AI still struggles to replicate: problem framing, relationship-building, insight generation, and navigating ambiguity. The message is blunt: don’t compete with AI on speed or memory; double down on the strengths only humans bring.

👉 Dec 15, 2025 — AEI Event: AI Governance: A Discussion with Rep. Jay Obernolte, featuring Rep. Kevin Hern

👉 Dec 17, 2025 — House Energy & Commerce Committee Hearing: Examining Biosecurity at the Intersection of AI and Biology

👉 Feb. 11–13 — The Complexities of AI in Health Care, AHLA, Las Vegas & Virtual

👉 Mar 12–18, 2026 — SXSW 2026, Austin, TX

👉 Mar 30–31, 2026 — IAPP Global Privacy Summit, Washington DC

👉 Apr 6–9, 2026 — HumanX 2026, San Francisco, CA

👉 Apr. 10 — Ethical AI: Leadership and Governance (Virtual)

And finally, if you like what you are reading, please share this newsletter with your networks and encourage them to sign up.

Till next time,

BC