Artificial intelligence can now prescribe mental health drugs

Artificial intelligence will see you now. 

Patients typically have to wait weeks and pay a hefty co-pay (if they have insurance) to get prescription renewals, but starting next month, the process will be quick and cheaper thanks to AI.

Legion Health, a Y Combinator-backed company that has raised $7 million since launching in 2021, will be the the first mental health program in the world with the authorization to allow AI to prescribe psychiatric medications. Initially, only patients in Utah will be able to use the feature, which requires a $20 per month subscription fee, but it plans to expand to other states soon.

“The long-term goal is to build the ‘AI doctor’ not as a black box that does everything, but as AI + doctors + clinic in the loop that can handle specific clinical tasks safely, transparently, and at scale,” Arthur MacWaters, 29, who cofounded Legion alongside Yash Patel, 30, and Daniel Wilson, 30, told NYNEXT. “The AI doctor thesis writ large has the potential to be one of the most valuable sectors on the entire planet.” 

The scope of the program is deliberately narrow for now. The AI can only renew what Legion deems “lower-risk psychiatric maintenance medications” — such as SSRIs, Wellbutrin, trazodone, and mirtazapine — that a human doctor has previously prescribed.

Patients explicitly opt into the AI system, which can be accessed via an app or web browser, and are clearly told they’re talking with an AI agent. After getting consent and verifying a patient’s identity, the AI conducts a two minute-long focused safety review covering drug interactions, side effects, and psychiatric warning signs.

Any red flag triggers immediate human takeover and patients can request human review at any point.

“The biggest risk is that they’re going to have to pay $300 out of pocket, drive two hours, and sit in an office for somebody who’s going to schedule them out two weeks to two months,” Wilson said.

In Utah, all 29 counties are designated as health professional shortage areas — indicating severe shortages in medical care. That means residents often can’t get basic prescription renewals without long waits and expensive visits.

“We genuinely believe that there aren’t enough human doctors on the planet to take care of all of the healthcare needs that there are,” MacWaters said. “AI is essentially critical.”

The pilot will follow a careful rollout: the first 250 prescriptions require doctor oversight, the next 1,000 get post-evaluation reviews by doctors, and only then does the AI operate autonomously.

Other companies have tried to disrupt healthcare with varying degrees of success — from Theranos’s blood testing fraud to Amazon’s failed Haven experiment to more successful ventures like Teladoc.

Doctronic, which was given the green light by Utah to prescribe more benign prescriptions like birth control has already proven troubled. Some researchers in a test were able to trick it into prescribing oxycontin and spreading vaccine misinformation, according to a report earlier this month.

Still, traditional healthcare is riddled with human mistakes — doctors juggling 30 patients a day and working with software that MacWaters described as feeling “like it came from a Terminator movie … this Windows 1994 stuff is crazy.”

AI doesn’t get tired, doesn’t forget patient history, and can review every page of someone’s medical records in seconds to catch drug interactions that overwhelmed human doctors might miss.

Utah has positioned itself as a national leader in AI policy, creating regulatory sandboxes that let companies temporarily bypass regulations to test new technologies. Unlike states that either ban AI or embrace it wholesale, Utah has chosen what officials call a “middle way” approach.

“We very much want to forge our own path,” Margaret Woolley Bussee, Executive Director of the Utah Department of Commerce, said. “We don’t want to be AI doom or AI boomer.”

Meanwhile, states like New York are more on the “doom” end of the spectrum, with proposed legislation that would ban AI systems like ChatGPT from answering any health-related questions—even basic queries about drug interactions or symptoms. The contrasting approaches highlight a growing divide between states that see AI as a solution to healthcare access problems and those that view it as too risky to allow in medical contexts at all.

Within a few years, crossing state lines could mean entering an entirely different healthcare universe—one where AI is either your medical assistant or completely forbidden. The technology might be advancing at breakneck speed, but politics will determine who gets to use it.

Regardless, Macwaters is bullish that AI in healthcare is here to stay. “Every patient is going to have AI working on their behalf in five years.”