Deputy Secretary of the Department of Health and Human Services Jim O’Neill is bringing a Silicon Valley attitude to AI use and regulation at the agency.
Along with “lighter regulation,” O’Neill expounded on how the federal government needs to provide regulatory clarity for AI companies and streamline the FDA approval process for medical devices. He expressed that he wants to guide policy towards building trust in devices, while tearing away at burdensome regulation.
O’Neill served in the Bush administration, then left D.C. for Silicon Valley, where he assumed he would stay. But when Trump was re-elected and tapped Robert F. Kennedy Jr. for health secretary, O’Neill believed he could make true changes to government.
“What was politically possible for HHS to do was radically transformed,” O’Neill said of his decision to rejoin the federal government. He spoke at an event on Wednesday hosted by Fire Arrow LLC and Nexus Policy Consulting.
O’Neill’s experience in Silicon Valley, and watching California’s legislative debates on implementing guardrails on technology, showed him that the government too often imposes impossible regulatory barriers that impede innovation.
“I’ve been involved in the periphery of AI, kind of, for the past 15 years in Silicon Valley, friends at all the big AI companies and a lot of the small ones as well. … There's been a debate in California also, basically pro-innovation or anti-innovation,” O’Neill explained. “And I've been on the pro-innovation side, not the safety side, for a long time.”
He continued: “AI can massively improve people's lives in every vertical but only if it's allowed to, and whether the forces holding it back are just strictly government or big companies trying to crush smaller companies. The Biden direction really seemed anti -innovation. … I think much lighter regulation is better for everyone.”
Though he expressed faith in the Health Insurance Portability and Accountability Act (HIPAA) for regulating patient data privacy and security, he said there are private sector solutions to increased security.
O’Neill said many AI companies have approached HHS offering solutions for government efficiency. He’s also had conversations with FDA-regulated device companies and those seeking reimbursement from CMS. All of them need clarity, he said.
“They all deserve regulatory clarity,” he said. “That's something that we need to provide. But some of them need FDA approval. Some of them need CMS reimbursement. And depending on what they are and what they're doing, we want to make sure that we're not standing in the way of anything.”
CMS is working on widening access to reimbursement for digital health technology and AI through its new ACCESS model, announced last week. FDA released a companion model called TEMPO, which will allow technologies in the program to initially skirt FDA requirements.
O’Neill decried that FDA has over-regulated wearables and clinical decision support technology. O’Neill explained that during his experience in Silicon Valley, health apps and wearables wanted to get FDA approval to be a medical device, but were deterred by the long, expensive and uncertain approval process.
Instead of pursuing FDA approval immediately, they would market themselves as consumer devices and avoid making health claims or interpreting data. After they’d generated some revenue or raised more capital, they would circle back to the pursuit of an FDA clearance.
“From the perspective of the patient that wants to use it, or, frankly, the doctor that wants to deploy it, it's a long, circuitous road where you pretend to be going one direction, even though your goal is over here. And, you know, it's regulatory clarity earlier. It would be great if people didn't have to pretend one thing in order to get to another place that's not the world that regulators should be creating.”
He predicted that every individual will have their own, personal AI agents that will help them manage their lives and “be loyal to you.” While he expressed some fear that AI agents would infringe on personal liberties, he thinks innovation is heading in the right direction.
“I think the future state … is that every human adult on earth has an AI agent that is loyal to and can help interface with others … take a lot of work off your plate as a human and help manage your life, help manage your data, help manage your resources, help out with whatever your particular weaknesses are, fill in the gaps, and most important, be loyal to you,” O’Neill said.
He continued: “I think we want to guide policy towards an area where you can trust your devices and your data and therefore get the benefit of getting health advice from a technology that you can trust.”
At the same event, Thomas Keane, M.D., the Assistant Secretary for Technology Policy and National Coordinator of Health Information Technology, said that he believes the Trump administration can figure out how to do post-deployment monitoring of AI, an unsolved issue for the FDA to effectively ensure safety of the technology.
O’Neill reiterated his distaste for the Coalition for Health AI (CHAI), which he said had positioned itself as a middleman between industry and government. He clarified that AI companies are welcome to engage directly with the federal agencies.
“Private companies have a right to team up with each other and say whatever they want to say,” O’Neill said. “But I did not want to leave an impression in people's minds that was the regulator, or that was the only way to talk to us. You could talk to us directly, and that's much easier and cleaner, and you don't pay dues to anyone.”
O’Neill has helped oversee the effort to give AI tools to employees of HHS, along with HHS’ Chief Information Officer and ex-Palantir executive Clark Minor.
HHS first deployed AI at the FDA in the spring, O’Neill said, where it has been used to help with summarizing and reviewing papers. In September, ChatGPT Gov was rolled out to the rest of the agency. Anthropic’s Claude became available last week, he said.
“The government is open for business,” CMS Administrator Mehmet Oz, M.D., told audience of virtual care company executives and telemedicine providers at the American Telemedicine Association’s EDGE policy conference hosted on Thursday.
It’s already inked contracts with OpenAI, Anthropic and Elon Musk’s x.ai. For a separate initiative, CMS’ Health Tech Ecosystem, it has entered into a contract with identity verification company CLEAR.
Within HHS, O’Neill said they are using models to detect “small signals” across its varied large datasets. He clarified that the agency is not combining datasets, but will share AI models across them.
AI has a massive role to play within HHS and in the healthcare industry more broadly, O’Neill emphasized.
”[I] have been excited to be able to roll out, so far, two large language models for all 65,000 of our employees, and working on using AI to improve medical care on the practice side, on the coordination care side, on the payment side, and also working to use it to improve reviews at FDA and Research at NIH and everywhere else.”