Cure All Disease in a Decade? Here's What AI Has Actually Delivered So Far

In early 2025, Google DeepMind CEO Demis Hassabis said something most scientists of his stature never say on camera: that AI could "cure all disease" within a decade. It's the kind of line that makes editors reach for exclamation points and makes actual biologists reach for their blood pressure medication. So we did what this publication is built to do — we went and checked what's actually been proven versus what's still a claim.

The honest answer is more interesting than either the hype or the dismissal: there is real, published, peer-reviewed acceleration happening in parts of the drug discovery pipeline. There is essentially no proven acceleration yet in the parts that matter most — safety, regulatory approval, and the biology of aging itself. Here's the evidence, layer by layer.

What's actually proven: protein structure, solved

Start with the one claim in this space that isn't contested. In 2024, the Nobel Prize in Chemistry was split between David Baker (computational protein design) and, jointly, Demis Hassabis and John Jumper for AlphaFold2 — the DeepMind system that predicts how proteins fold into their 3D shapes. The Nobel committee's framing was specific: AlphaFold made it possible to predict structures for essentially all 200 million proteins known to science, a problem that had resisted brute-force experimental methods like X-ray crystallography and cryo-EM for roughly 50 years.

This is real and it matters, because protein shape determines protein function, and function is the thing drug developers are actually trying to target. But it's worth being precise about what AlphaFold did and didn't do: it solved a structural prediction problem, not a "cure disease" problem. Knowing a protein's shape is a necessary input to drug discovery — it is not a drug.

What's in actual human trials right now

The clearest test of the "AI drugs are coming" claim is Insilico Medicine's rentosertib (also known as ISM001-055/INS018_055), a TNIK inhibitor for idiopathic pulmonary fibrosis (IPF) — and it's a genuinely notable data point, because the drug's target and initial design were both generated by AI models rather than discovered through traditional screening.

The phase 2a trial results were published in Nature Medicine in 2025: a randomized, double-blind, placebo-controlled study that reported the drug was safe, well-tolerated, and showed encouraging signals on lung function measures versus placebo over 12 weeks. That's a legitimate milestone — it's one of the first AI-originated drug candidates to clear a controlled human efficacy trial with a positive readout.

It is also, notably, a phase 2a trial in one indication. It is not evidence that AI has compressed the traditional 10-year, multi-billion-dollar drug development timeline — it's evidence that one AI-discovered molecule cleared one early efficacy hurdle. Phase 3 trials, the ones regulators actually weight for approval, are still ahead of it.

The animal proof of concept: dogs, not humans

The most concrete "aging drug" story right now isn't happening in humans at all. Loyal, a biotech startup, has been running LOY-002 through the FDA's veterinary approval pathway for senior dog lifespan extension. As of the most recent updates, the drug has cleared the FDA's "reasonable expectation of effectiveness" (RXE) determination and completed the safety package — two of three requirements on the path to conditional approval, which the company is targeting for 2026.

This is worth taking seriously precisely because dogs develop many of the same age-related conditions humans do — cancer, cognitive decline, organ deterioration — making canine trials a genuinely useful proxy. But "reasonable expectation of effectiveness" is a regulatory threshold for conditional approval, not proof of efficacy, and none of this data has been shown to translate to humans. Treat it as the most credible early signal in the field, not as a preview of a human longevity pill.

The N-of-1 in the room: Bryan Johnson

No conversation about AI and longevity happens online anymore without Bryan Johnson, who has turned his own body into a continuously monitored experiment under his "Blueprint" protocol. His claims — a 5.1-year reduction in biological age over two years, a self-reported "speed of aging" of roughly 0.64 (about 7.5 months of biological aging per calendar year) — get cited constantly as evidence the aging problem is cracking.

It isn't evidence of anything beyond what happened to one well-resourced individual running dozens of simultaneous interventions with no control group. Even clinicians featured in Johnson's own Netflix documentary make this point: when you change diet, sleep, supplements, exercise, and monitoring protocols all at once, you cannot isolate what's actually driving any measured change. Johnson is a fascinating personal experiment. He is not a clinical trial, and the aging-clock literature doesn't treat him as one.

The real, modest human data

The most methodologically solid human aging data we found doesn't come from a biohacker or an AI lab — it comes from the DO-HEALTH trial, published in Nature Aging in 2025, which found that 1 gram of daily omega-3 supplementation over three years produced a small but statistically measurable slowing of biological aging across several epigenetic clocks.

"Small" is the operative word. The effect was measured in months of biological age difference, not years, and the study's own authors caution that movement on an aging clock is not proof of functional age reversal. This is what real, peer-reviewed, controlled longevity research currently looks like: modest, slow, and honestly reported. It is the opposite of a decade-to-cure-everything headline.

So what did AI actually change?

Laid side by side, the evidence supports a narrower and more defensible claim than "AI will cure disease in ten years." What's actually happened is that AI has meaningfully compressed one specific layer of the drug development stack: computational discovery and design — identifying targets, predicting structures, generating candidate molecules. AlphaFold's Nobel Prize and rentosertib's phase 2a readout are both real evidence of that.

What AI has not yet touched, in any demonstrated way, is the layer that actually determines how long drug development takes: safety testing, large-scale efficacy trials, regulatory review, and manufacturing scale-up. Hassabis's own "months instead of years" framing was specifically about the design phase — the "cure all disease" leap is a much bigger claim than the data he's actually sitting on supports today.

The concept researchers call "longevity escape velocity" — medicine improving the body faster than it declines — remains a coherent hypothesis, not a demonstrated trajectory. The building blocks people point to (senolytics, partial cellular reprogramming, precision gene editing) are real research programs with real early-stage results. None of them have produced human outcome data suggesting the curve has actually bent.

The verdict

This is a genuine scientific story, not a manufactured one — AI-originated drugs are now clearing controlled human trials, and that's new. But the distance between "AI compressed the discovery phase of one antifibrotic candidate" and "AI will cure all disease within a decade" is enormous, and it's exactly the distance where hype cycles live. Watch rentosertib's phase 3 data and Loyal's conditional approval timeline — those are the two most concrete, trackable tests of whether this acceleration actually reaches patients. Everything else, for now, is a very well-funded thesis.