Why David Sacks’ appointment could grease the wheels of medical AI innovation

For Avandra CEO and founder Ryan Tarzy, the potential of AI in healthcare is personal. He lost his father to pancreatic cancer, a disease often caught too late in the United States. At the recent Radiological Society of North America (RSNA) conference, he encountered a telling example of what emerging medical tech can achieve. “At RSNA, I saw AI that can detect pancreatic cancer at one centimeter,” he explained. “Humans usually catch it at two centimeters or more, and by then, survival odds drop drastically. Early detection can save lives, and that’s where AI excels.”

This kind of early detection is not just theoretical. In Taiwan, for instance, National Taiwan University Hospital announced earlier this year the development of an AI-assisted model capable of spotting pancreatic tumors smaller than 2 cm—significantly earlier than most human specialists. Research is beginning to document these capabilities rigorously. A recent nationwide population-based study in Taiwan, published in Radiology (Chen and Wu et al, 2023), found that a deep learning–based tool could detect pancreatic cancer on CT scans with 89.7% sensitivity overall. The system achieved 74.7% sensitivity for detecting tumors smaller than 2 cm—precisely the size threshold at which early intervention can drastically improve survival rates. This stands in contrast to conventional detection methods, which can miss up to 40% of tumors under 2 cm.

Despite the existence of such tools overseas, U.S. regulations stand in the way, Tarzy said. “It’s harder to gather that data here because we don’t detect it early often.” This regulatory conundrum, he believes, slows the arrival of lifesaving tools.

“There’s AI overseas that can detect it earlier, but FDA rules require significant U.S.-based data for approval. Without early-detection tools, it’s tough to generate that data. This creates a frustrating catch-22: the FDA requires U.S.-based data to approve early-detection tools, but without those tools, we struggle to generate the early-detection data needed for approval. Why not accept data proven effective elsewhere? Such regulatory inflexibility delays lifesaving innovation.”

Now, with President-elect Donald Trump appointing tech entrepreneur and venture capitalist David Sacks as the White House AI and crypto “czar,” Tarzy and others see a potential shift. “[Sacks] has spoken about the danger of regulatory capture—where only large corporations can navigate stringent regulations—thus stifling startups and mid-sized innovators,” Tarzy said. “Without balanced policy, we risk ending up with just a few big players controlling foundational models.”

Breaking down data barriers

Avandra, which emerged from stealth in October 2024, aims to solve a fundamental data-access problem. Using a federated network model, the company offers secure, de-identified imaging data from around the world—potentially tapping into portions of the estimated 2 trillion medical images globally. Hospitals can earn revenue from their imaging assets while pharmaceutical companies, AI developers, and researchers gain access to the datasets they need. Avandra thus aims to extend the life of medical images beyond their initial diagnostic use.

It’s a timely mission. According to Tarzy, many hospitals are technologically stuck in the past. “Healthcare is lagging technologically,” he said, describing an environment still reliant on “Windows seven, legacy .NET frameworks, pagers, fax machines, and even burning millions of CDs a year.” At RSNA, an NVIDIA representative noted that “healthcare is about three technological generations behind” other industries, Tarzy recalled. “That’s probably not wrong. It’s certainly not true in every setting of healthcare, but it’s common enough.”

Avandra’s federated approach aims to bring medical imaging into the present—enabling faster AI training, more streamlined innovation, and ultimately, better patient outcomes.

A potential regulatory inflection point

The U.S. healthcare sector’s slow embrace of transformative technology isn’t just technical—it’s also regulatory. “We were trending toward heavier regulation, and I get the safety concerns,” Tarzy acknowledges. “But if we over-regulate too soon, we halt progress that could save lives. We’re in early AI days, and regulators don’t yet fully grasp the technology’s nuances.”

Sacks’ appointment offers hope for a more innovation-friendly environment. “He might push for more agile frameworks that let good AI through faster, balancing safety with innovation,” Tarzy said, contrasting the current landscape with Europe’s more restrictive model. Taiwan’s early-stage detection AI and its FDA breakthrough designations show what’s possible when regulatory pathways are navigable. “The future is here, but it’s not evenly distributed,” Tarzy noted, referencing sci-fi author William Gibson’s quote. Under Sacks, the U.S. could realign policies that currently limit life-saving advances.

From replacement to augmentation

Years ago, luminaries like Geoffrey Hinton and Vinod Khosla predicted that AI would replace radiologists. In 2016, Hinton likened radiologists to Wile E. Coyote from the cartoons, stating, “If you work as a radiologist, you are like Wile E. Coyote in the cartoon; you’re already over the edge of the cliff, but you haven’t looked down yet.” Similarly, in 2017, Khosla said radiologists would be “obsolete in five years.” But AI hasn’t replaced radiologists — who continue to be in short supply; it’s assisting them, handling repetitive tasks and highlighting subtle findings, such as early-stage pancreatic tumors, that humans might miss.

Rather than making radiologists obsolete, the opposite has occurred: radiologists are busier than ever, with growing imaging volumes and persistent shortages. AI can serve as a co-pilot, extending human capabilities rather than eliminating skilled professionals.

“The volume of imaging keeps growing, and there aren’t enough radiologists to meet that demand,” Tarzy emphasizes. “AI is critical—it can handle repetitive tasks, highlight subtle findings, and let radiologists focus on complex interpretation.”

Data, policy, and the future of healthcare

Ultimately, Avandra and companies like it exist at the intersection of data availability, regulatory flexibility, and technological advancement. “Medical imaging is the ‘camera’ of healthcare AI,” Tarzy said. By unlocking imaging data and aligning policies with modern realities, innovators can accelerate the availability of tools that detect diseases earlier and improve outcomes.

“It’s a big systemic issue: fee-for-service models, limited value-based care, socioeconomic factors, dietary issues, and outdated reimbursement policies,” Tarzy said, acknowledging that regulatory reform alone isn’t a cure-all. But it can pave the way for progress. With more flexible policies and better data, life sciences and AI innovators can bring powerful tools to market faster.

A forward-looking regulatory mindset

Sacks’ entrepreneurial track record suggests he understands how foundational AI models evolve—and how startups must adapt. In a June interview with Axios, he spoke to the sink-or-swim dynamic that foundation models pose to startups.

“The key for apps is, as those foundation models become more and more powerful, does that help you or hurt you?” Sacks said. “If what [a company] had spent the last two years doing was conversation, they just became obsolete… You have to be very, very careful when you’re building on top of these models to understand where their innovation ends and your innovation begins.”

Such realities resonate in medical AI, where companies like Avandra must navigate both evolving AI capabilities and complex regulatory requirements. If Sacks can help shape policies that recognize this dual challenge, it could break the current regulatory logjam: enabling companies to build upon powerful AI advances while maintaining the rigorous safety standards healthcare demands. For patients waiting on earlier detection tools already proven effective overseas, such a shift can’t come soon enough.