The Bandwidth - Weekly AI Roundup

The competition for dominance at the intersection of AI and health accelerated again this week.

Recall that last week, OpenAI published research on how people already use ChatGPT for health-related questions. It also rolled out two new offerings. The first, ChatGPT Health, is consumer-facing. The second, OpenAI for Healthcare, targets health systems with secure, HIPAA-aligned tools, including ChatGPT for Healthcare, now deploying across large providers such as AdventHealth, HCA Healthcare, Memorial Sloan Kettering, and UCSF.

This week brought more revealing moves about the company’s intentions around health. OpenAI announced it is acquiring Torch, a startup focused on creating a “unified medical memory” that pulls fragmented patient data into a single, usable layer for AI systems. Hats off to the Torch team if they have cracked the code on unifying our health data. The reported price: $60 million. The team: four people. That’s right, four people. Which gives some credence to prognosticators’ who say AI could lead to many more microbusinesses of significant value.

Not to be outdone, Anthropic announced Claude for Healthcare, also offering HIPAA-ready tools for providers, payers, and consumers to use Claude in medical contexts. It also expanded its life sciences capabilities, building on work introduced last fall, with deeper connections to scientific platforms and stronger support for clinical trials, regulatory operations, and regulatory workflows.

Two questions now loom, IMO.

First, these companies are clearly making large, long-term bets on healthcare, often through partnerships with both legacy institutions and nimble startups. But how well their models can deliver reliable, safe medical guidance directly to consumers remains an open question. Skepticism is already high, and legal scrutiny is not hypothetical. It is underway.

Second (and less discussed), on the enterprise side, the challenge is different. These are not healthcare-native companies. As they move deeper into B2B offerings, they will be competing with incumbents that have decades-long customer relationships, deep operational trust, and their own AI capabilities increasingly bolted onto existing platforms. Whether technical superiority is enough to displace that kind of institutional gravity is far from settled.

One thing frontier labs like OpenAI or Anthropic have going for them: aggregation. As was pointed out in this Axios article about the role of AI in pharma development, “Those platforms often generate stronger insights by aggregating data across multiple pharma partners …. Everybody benefits from the common algorithms.”

Apple taps Google’s AI to power the next version of Siri
Apple’s multi-year deal with Google, alongside its continued OpenAI partnership, points to a multi-model orchestration strategy where platform owners decide which AI runs where.

Google launches “Personalized Intelligence”
With the user’s permission, Google’s new feature connects Gemini to your Gmail, Photos, Search, and YouTube history to give AI responses tailored specifically to you. This is a clear demonstration that Google’s set of apps and large user base may prove to be its strongest moat against rivals.

NVIDIA and Eli Lilly launch a $1B AI lab for drug discovery
The partnership signals pharma’s shift from pilots to capital-intensive AI infrastructure, treating AI as core R&D, not an efficiency add-on.

AI-driven programs expand to address physician shortages
Programs focused on augmenting clinical capacity, not replacing clinicians, reflect a pragmatic adoption path amid workforce constraints.

Google pulls AI Overviews for some medical queries
A reminder that health remains a high-risk domain where confidence, liability, and trust constrain automation.

AI skin-aging prediction moves onto a wearable patch
MIT and Amorepacific’s ultra-thin sensor predicts skin aging using AI.

FDA and EMA align on “Good AI Practice” principles
International regulators are aligning on a common set of principles to guide the responsible use of AI across the full drug product life cycle, from discovery and clinical trials to manufacturing and post-market surveillance. The goal is to harness AI’s potential to speed development, improve safety monitoring, and reduce reliance on animal testing, while reinforcing core regulatory standards around quality, safety, and efficacy. The two-page document lays out 10 high-level principles and calls for global collaboration on research, standards, education, and regulatory harmonization as AI use in drug development continues to evolve.

Trump vows to make Big Tech cover AI data center power costs
President Donald Trump said U.S. tech leaders will make "major changes" to ensure that Americans will not face higher bills for power used by artificial intelligence (AI) data centers. "I never want Americans to pay higher electricity bills because of data centers," Trump wrote in a post on his social media platform Truth Social on Monday. The truth is the industry may be moving more quickly than he can to address this public concern (more on that below).

CBO credits GDP growth to generative AI
Last week (after we published), the Congressional Budget Office released its view of the economy through 2028. Of note, the actuaries stated, “From 2027 to 2028, a mix of positive and negative factors leads to average real GDP growth of 1.8 percent per year. In those years, growth is supported by increases in the labor supply and in investment that result from the 2025 reconciliation act and by the positive effects on productivity stemming from the adoption of generative artificial intelligence.”

Microsoft pledges “community-first” AI infrastructure
Microsoft rolled out a “Community-First AI Infrastructure” initiative aimed at preempting the backlash around the massive data centers required for AI. The plan rests on five commitments: the company will pay its own electricity costs so local rates do not rise, minimize and replenish water use, create local jobs, contribute to local tax bases, and invest in community AI training and nonprofit support. The underlying logic is that AI infrastructure should strengthen host communities rather than strain them, and that responsible build-out requires early engagement, transparency, and collaboration with utilities and policymakers.

AI nutrition labels gain traction as a governance tool
This piece argues that “responsible AI” in healthcare must be grounded in formal oversight, robust governance, transparency, and unwavering commitment to patient safety and regulatory compliance. It emphasizes involving clinicians early and continuously so that tools address real workflow friction rather than adding burden, with a focus on targeting low-value administrative tasks first. Risk assessment should determine how much transparency and monitoring a model requires, and tools must be accompanied by clear documentation, like model cards - akin to nutrition facts labels for AI.

Microsoft: Global AI adoption rises, but gaps widen
Microsoft’s AI Economy Institute reports Global AI adoption kept climbing in late 2025, but leadership is concentrating, not spreading. The UAE is now the clear front-runner, with 64 percent of its working-age population using AI, extending its lead over Singapore and the rest of the field after years of heavy investment in digital infrastructure, skills, and government use. The U.S., by contrast, is slipping: despite leading in AI research and infrastructure, it fell to 24th globally in adoption, with just 28.3 percent usage, trailing behind smaller, more digitally coordinated economies. In short, building the best AI does not guarantee that people actually use it.

Prompt tip: turn AI into an on-demand think tank
From our friends at the Nueron, here’s a guide on how to use structured prompts to create your own think tank or run a red team/blue team exercise. Scroll down to the Jan 13 entry.

Claude Cowork brings agent-style work to non-technical teams
Anthropic introduced Cowork, a research-preview feature for its Claude AI that brings agent-style work into everyday productivity tasks. Unlike typical chat interactions, Cowork lets users grant Claude access to a specific folder on their computer so it can read, create, and edit files autonomously, plan multi-step tasks, update the user on progress, and integrate with external connectors for more complex workflows. It aims to make Claude feel more like a digital coworker than a simple chatbot. According to some reviewers, here’s what Cowork can already accomplish:

Watch the video below from the Anthropic team to learn more.

👉 Jan. 20 — Northwestern’s Kellogg School of Management Webinar: The Insightful Leader Live - How AI Reflects Human Bias … and What to Do About It

👉 Jan. 21 — AI in Nutrition Evidence Assessment by the National Academies.

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

👉 Feb. 23 — Reminder: HHS seeks input on AI in clinical care by this date

👉 Feb. 26 - Harvard Business Review Strategy Summit: Build your transformation playbook for the AI era.

👉 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. ✍️ 🆙 And/or, give me a shout out on LinkedIn.

Till next time,

BC