A Utah startup is now running an AI that autonomously renews prescriptions for chronic conditions. After a physician signs off on the first 250 renewals, the AI takes over.

Antidepressants, proton pump inhibitors, and blood pressure meds, all with no oversight or any human in the loop after that.

It's operating under Utah's "AI sandbox," a state framework that grants temporary waivers from state law to test medical AI tools in the public. It’s the first time I’ve heard of this setup, but the real problem is that federal law (the Food, Drug, and Cosmetic Act) likely requires a licensed human "prescriber" to authorize any refill.

So this pilot may be illegal at the federal level. So far, the FDA has said nothing.

And THAT is where the bigger story is, because why aren’t they? We're watching kind of regulatory devolution, where states build the sandboxes, startups operate inside them, and the FDA chooses not to act.

Inaction sets policy just as surely as a rule would, and the longer it goes on, the more it normalizes. So, how long until the sandbox just becomes the standard?

For more details: Full Article

Algorae signed a deal with Zydus to exclusively commercialize portions of its portfolio, a major step in turning its AI-discovered drugs into revenue. [Source]

Insilico, despite receiving clinical trial approval for its Rentosertib inhalation formulation on April 28, is drifting lower. [Source]

Select the right answer! (See explanation below and source)

iRegene Therapeutics just registered a Phase 1 trial of NouvNeu001, an AI-designed cell therapy, in Parkinson's patients who have already failed deep brain stimulation or lesioning surgery. The 10-patient single-injection study is set to start in May 2026 and targets a notoriously difficult population whose disease has progressed past existing surgical options. [Link]

MindRank just posted two new Phase 1 clinical pharmacology trials for MDR-001, its AI-designed oral GLP-1 agonist that entered Phase 3 in February. The studies, both starting in May or June, will characterize drug-drug interactions with the CYP3A4 inducer rifampin and inhibitor itraconazole (NCT07550621), and pharmacokinetics in patients with mild and moderate hepatic impairment (NCT07550634).

Trump fired the entire National Science Board without formal announcement, with members learning via termination emails. The move comes alongside an FY2027 budget proposal that would cut NSF funding by 54% and reduce AI research funding from $965M to $655M. NSF funds a significant share of the academic computational biology and AI-for-science research that feeds the commercial AI drug discovery pipeline. [Link]

The FDA just launched a pilot to review clinical trial data in real time, starting with cancer trials from AstraZeneca and Amgen. AstraZeneca is running a Phase 2 lymphoma trial at MD Anderson and UPenn, and Amgen is running a Phase 1b small cell lung carcinoma trial, both monitored on a platform built by Paradigm Health. The agency simultaneously opened a public comment period on a broader AI pilot covering safety monitoring, dose selection, safety signal identification, and patient recruitment. [Link]

Johnson & Johnson says AI has cut the time to prepare a clinical trial report from 700 to 900 hours down to about 15 minutes. CIO Jim Swanson disclosed the figure at the Reuters Momentum AI event, adding that J&J has also halved its lead optimization time for new drug candidates. Swanson cautioned that "discovering new products outright and bringing them to market using AI is not yet possible." [Link]

A team at GATC Health and UC Irvine just published some very interesting findings in PNAS. They took genetic and chemical signals from the brains of people who had died with opioid addiction, fed it into an AI platform, and asked the model to figure out where the dysfunction was and what might fix it.

The AI didn't point them toward opioid receptors. It pointed them toward serotonin, the same brain system classic psychedelics like psilocybin and LSD work on.

The model then sifted through more than 80 candidate molecules and picked a winner.

GATC-1021 is a small drug designed to switch on two specific serotonin receptors, with a chemical structure that echoes a 2021 compound built to deliver the brain-rewiring benefits of psychedelics without the actual trip. The AI's prescription for fentanyl addiction, in other words, was build a psychedelic that doesn't make you hallucinate.

In rats trained to press a lever for fentanyl injections, GATC-1021 cut drug-seeking by more than 60%, and the effect held up across five days of dosing. The drug reached the brain, the rats didn't do the head-shake that rodents do when they're tripping, and the tiny knobs brain cells use to form new connections (called dendritic spines) sprouted in larger numbers in the hippocampus.

That kind of physical rewiring is exactly what the psychedelic field has come to believe is the real reason psilocybin can lift depression. The genetic machinery for forming new connections also lit up in the prefrontal cortex.

Every approved drug for opioid addiction works on the opioid system itself. Methadone and buprenorphine plug into the same receptors fentanyl does. Naltrexone blocks them.

None of these drugs touches the underlying brain rewiring that makes addiction so hard to escape. GATC-1021 ignores opioid receptors entirely and aims at the rewiring directly. It also lands in arguably the hottest area in brain drug development right now.

Delix Therapeutics, spun out of UC Davis, got FDA clearance last October for at-home dosing of a similar non-hallucinogenic psychedelic in depression, plus a federal grant to test it in addiction.

Psilocybin itself is in addiction trials at Wisconsin, Penn, Johns Hopkins, and Yale. Serotonin-driven brain rewiring is becoming the field's bet on what direct opioid drugs can't fix.

Eleven of the paper's authors work for GATC, including the chief scientist, the chief technologist, and the principal AI engineer. The company has a partnership with Lisata Therapeutics aimed at human trials, which Lisata said in June would start in early 2026.

It's the end of April, and no submission to regulators has been announced yet. The doses tested in rats were very high. And a vitamin-derived supplement the AI insisted on pairing with the drug, supposedly to help it reach the brain, actually did the opposite.

It blocked brain absorption and produced its own modest effect on fentanyl-seeking, which muddies the cleanest version of the story. There's a deeper question lurking too: if the AI lands on a class of compounds human chemists were already chasing, what is it adding besides speed?

Plenty of non-hallucinogenic psychedelic candidates are already racing toward addiction and depression markets, most of them designed by humans.

Sources: [Research Article]

Metformin is generally recommended as the initial pharmacologic therapy for Type 2 diabetes due to its effectiveness, safety profile, and low cost. GLP-1 receptor agonists are also widely used, particularly when additional glucose control or weight loss is desired, but they are typically added after or alongside metformin in many guidelines. [Source]