Trump rolls back artificial intelligence order

Following President Trump's inauguration, our eyes are on sweeping changes he's pushed through executive orders, and as expected, he rolled back former President Biden's order on artificial intelligence from 2023. At the time, STAT's Casey Ross did a rundown of what the Biden EO meant for health AI developers. The best-known provision required developers of foundation models "that may pose a risk to national security" to provide information to the federal government, including about their safety testing. This was incredibly unpopular with industry.

The order also made a number of demands of the Department of Health and Human Services, including that it develop a strategic plan for artificial intelligence, which it only just delivered, and an inventory of ways the department makes use of AI. Despite the recent excitement around AI, the federal government and Congress did not make much progress on guardrails around the technology, in health care or otherwise, though regulators made some limited rules. It's unclear what exactly the Trump administration will do, but i t's certain to be a more hands-off approach.

resarch

A look at marketing for compounded GLP-1s

GLP-1s are the hottest drugs in a generation, and if you have eyeballs, you've been bombarded with marketing about the popular obesity and diabetes drugs' tremendous benefits. But are you sure everything you're reading online is true?

The drugs are expensive, demand has led to shortages, and online services have rushed to fill the void with lower-cost compounded copies. As STAT's Katie Palmer reports, a new study found that of 79 websites marketing compounded GLP-1s or prescriptions for them, 37% stated or implied that the drugs were FDA-approved. Nearly half didn't include information about the drugs' adverse effects, warnings, and contraindications. Compounded copies are not approved by FDA.

The study underscores concerns about the proliferation of compounded GLP-1s, which may or may not be on the down swing as shortages ease and as drug companies consider ways to offer the brand-name drugs at lower cost.

Model card madness

Screenshot 2025-01-21 at 7.25.32 AM

Following a federal deadline at the end of the year, developers of certain AI products used in health care must provide end-users a number of details about how they were developed and how they should be used. The use of "model cards," or "nutritional labels," to detail machine learning and other technologies is widespread in the broader tech industry, but its use in health care is newer. So to help developers comply with the new rules, or simply to help them disseminate information about their products, the Coalition for Health AI and the Duke Institute for Health Innovation have each published free-to-use templates.

Interestingly, though the Food and Drug Administration hasn't actually mandated that developers of regulated AI medical devices detail the products with model cards, its draft guidance document for AI developers released earlier this month says model cards can be useful and includes an example. Among the concerns of industry is that dueling regulatory regimes will make compliance difficult, so it will be interesting to see what efforts are made to harmonize recommendations across the federal health department.

Medical devices

WuXi AppTec unloads med device testing unit

Brittany Trang writes: Last week, China-based WuXi AppTec sold its U.S. medical device testing operations to NAMSA, a private-equity-owned American company. The move is the latest in a string of WuXi divestments after the BIOSECURE act passed the U.S. House last fall. That law would have prohibited pharma and biotech companies from using equipment or services from foreign companies "of concern," largely understood to be Chinese companies like WuXi AppTec and WuXi Biologics. However, the legislation got left out of year-end packages and will likely be harder to pass under the new administration. As part of the deal, NAMSA will take over WuXi's sites in Minnesota and Georgia.

science

EvolutionaryScale's new protein model

Brittany also tells us: EvolutionaryScale, a protein AI company founded by employees from Meta's former protein AI division, last week published a new protein language model that can process not only the sequences of proteins, but also their structures and functions. The model, ESM3, was previously disclosed as a preprint last summer co-authored by researchers from the Arc Institute. To show that the model can preserve function while producing sequences that are different than known ones, researchers used it to produce a version of green fluorescent protein that was only 58% similar to other versions. EvolutionaryScale aims to provide protein design AI technology to other companies and raised $142 million in a 2024 seed round with participation from Amazon and NVentures, chipmaker NVIDIA's venture capital arm.

Want more health AI news? Sign up for Brittany's AI Prognosis newsletter launching in February.

More around STAT

Check out more exclusive coverage with a STAT+ subscription

Read premium in-depth biotech, pharma, policy, and life science coverage and analysis with all of our STAT+ articles.

What we're reading

Inside the looming crisis in the infant formula market, STAT
Online Therapy Boom Has Mainly Benefited Privileged Groups, Studies Find, New York Times
Sutter Health strikes deal with GE HealthCare for AI imaging tech as it maps out ambitious expansion plans, Fierce Healthcare
Trump orders U.S. to withdraw from World Health Organization, STAT