In mid-to-late June 2026, one of the most important scientists of his generation quietly changed jobs — and it tells you where AI is headed next. John Jumper, who shared the 2024 Nobel Prize in Chemistry for using AI to solve a 50-year-old biology problem, announced he is leaving Google DeepMind after nearly nine years to join Anthropic, the company behind Claude. When a Nobel-winning structural biologist leaves the lab that made him famous to join a large-language-model company, that is not a normal hire. It is a signal that the frontier AI labs believe their next big prize is not a better chatbot — it is science and medicine. Sources: TechCrunch, June 20, 2026; CNBC, June 19, 2026.
This is the honest guide to what that means. It is easy to get swept up in headlines that AI is about to cure cancer, and just as easy to dismiss it all as hype. The truth in 2026 is more interesting than either: AI has produced a handful of genuinely real medical milestones, and it has also run headfirst into hard limits that keep a human doctor firmly in charge. Here is what has actually happened, what has not, and what it means for you the next time you are tempted to ask an AI about a symptom.
Quick summary: John Jumper — co-winner of the 2024 Chemistry Nobel for AlphaFold, the AI that predicts protein shapes — is leaving Google DeepMind for Anthropic (his exact role there was not disclosed). His move lands during a real wave of AI-in-medicine progress: the first AI-designed drug (Insilico's rentosertib) posted positive Phase II results and entered Phase III, and the FDA has now authorized 1,451 AI-enabled medical devices. But the limits are just as real — no fully AI-designed drug has FDA approval yet, medical AI still hallucinates, and on a tough new biology benchmark the best model scores only about 31%. AI is becoming a powerful tool for doctors and scientists, not a replacement for them.
Who Is John Jumper, and Why Does This Matter?
For 50 years, one of biology's hardest problems was predicting how a protein folds. A protein is a chain of amino acids that scrunches into a specific 3D shape, and that shape determines what the protein does — and what drugs can bind to it. Figuring out a single structure in a lab could take years of painstaking work. Jumper led the DeepMind team that built AlphaFold, an AI system that predicts a protein's structure in minutes instead of years. The AlphaFold database now holds more than 200 million predicted structures, covering nearly every protein known to science, and researchers in 190 countries use it. In 2024, Jumper shared the Nobel Prize in Chemistry with DeepMind CEO Demis Hassabis for that breakthrough, with the other half going to David Baker for protein design. Sources: Google DeepMind, October 2024; AlphaFold DB / EMBL-EBI.
So why does it matter that he is joining Anthropic? Because it is a defection at the exact spot where AI and biology meet. Anthropic did not disclose what Jumper will work on, and it would be wrong to invent a title for him. But you do not hire the person who cracked protein folding to answer customer-support emails. The move is the clearest sign yet that the labs racing to build the smartest general AI now see their models as instruments of scientific discovery — and that the talent war has moved from who can build the best chatbot to who can build the best scientist. Reportedly, another senior Google DeepMind figure — Gemini co-lead Noam Shazeer — left for OpenAI the same week, underscoring how fierce that competition has become. Sources: TechCrunch, June 20, 2026.
What AI Has Actually Done in Medicine (The Real Wins)
Strip away the hype and there is a real, verifiable list of things AI has accomplished in medicine over the past year. These are not promises; they are results with data behind them.
- The first AI-designed drug reached late-stage trials. Insilico Medicine's rentosertib — a treatment for idiopathic pulmonary fibrosis, a scarring lung disease — was both discovered and designed using AI. In results published in Nature Medicine in 2025, patients on the highest dose (60 mg once daily) improved lung function by about 98 mL, while those on placebo declined by roughly 20 mL, and it has since advanced to Phase III trials. It is the strongest clinical signal to date that an AI-designed molecule can show real benefit in humans.
- AI is already inside the clinic. As of the end of 2025, the FDA had authorized 1,451 AI-enabled medical devices, with a record 295 cleared in 2025 alone. About three-quarters are in radiology — tools that help spot tumors, fractures, and bleeds on scans faster and flag what a radiologist should look at first.
- Serious money is betting on AI drug discovery. Isomorphic Labs, a DeepMind spinout, raised $2.1 billion and is preparing its first human trials of AI-designed cancer drugs. The scale of investment reflects genuine scientific confidence, not just hype.
- The big AI labs have moved in. Anthropic launched Claude for Life Sciences in late 2025 and a research workbench called Claude Science in mid-2026, and struck a deal with pharma giant Bristol Myers Squibb. OpenAI released GeneBench-Pro, a benchmark for genomics reasoning. Biology is now a first-class target for frontier models.
Read together, these say something real: AI has crossed from 'interesting demo' to 'produces results that pass peer review and clear regulators.' That is a meaningful line to have crossed, and it is why a scientist of Jumper's caliber is making this bet. Sources: Nature Medicine via EurekAlert, June 2025; FDA AI device list, 2025; Forbes, May 2026; CNBC, June 30, 2026.
What AI Still Can't Do (The Hard Limits)
Now the other half of the truth, because leaving it out would be exactly the kind of hype this guide is trying to avoid. For every real milestone, there is a hard limit that keeps AI a tool rather than a replacement.
First, no fully AI-designed drug has FDA approval. Positive trial results are a huge deal, but the road from Phase II to an approved medicine is long and littered with failures; rentosertib is promising, not finished. Second, medical AI still hallucinates. Studies through 2025 show that even models specialized for medicine can be pushed into confidently stating false things under messy or adversarial input — which in a clinical setting can be dangerous. Third, AI is nowhere near a research-grade biologist. On OpenAI's own GeneBench-Pro test of genuine genomics reasoning, the best model scored only about 31%, and experts estimate each problem takes a human specialist 20 to 40 hours — a blunt reminder of how far models are from replacing scientific judgment. Sources: Nature Communications Medicine, 2025; OpenAI GeneBench-Pro, June 2026.
You will also see headlines that AI "beats doctors" at diagnosis. Handle those carefully. Microsoft's much-discussed diagnostic system did hit 85.5% on a set of tough medical cases versus about 20% for physicians — but the doctors in that test were forbidden from using textbooks, colleagues, or the internet, which is not how any real doctor works. Strip away the artificial handicap and the honest claim is narrower: AI can be a superb second opinion and pattern-spotter, not a substitute for a clinician who examines you, knows your history, and is accountable for the outcome. Sources: Time, June 2025; Fortune, July 2025.
What This Means for You
For everyday life, the practical takeaways are simple. AI is becoming an excellent tool for understanding your health — explaining a diagnosis in plain English, helping you prepare questions for an appointment, translating a wall of lab results into something readable. It is not a doctor, and it should never be the last word on a medical decision. Use it to get informed and to ask better questions, then take those questions to a professional who can examine you and is responsible for your care. If an AI answer about your health feels alarming or certain, that is exactly the moment to verify it with a human.
The bigger picture is genuinely hopeful. The same technology powering the chatbots you use every day is starting to design medicines, read scans, and accelerate the science that leads to cures — and the people who understand it best, like John Jumper, are moving to where they think that work will happen fastest. If you want to explore this responsibly, a multi-model tool like LumiChats lets you put the same health question to several top models at once — useful for turning a diagnosis into plain English and better questions for your doctor, not for reaching a verdict. Where the models disagree is a signal of how much to verify with a professional, never a substitute for one.
01Did John Jumper really leave Google DeepMind for Anthropic?
Yes. In mid-to-late June 2026, after nearly nine years at DeepMind, most of it leading AlphaFold, Jumper announced he is joining Anthropic following a short break. Neither company disclosed his specific role there.
02What did he win the Nobel Prize for?
He shared the 2024 Nobel Prize in Chemistry with DeepMind CEO Demis Hassabis for AlphaFold, an AI that predicts protein structures. The other half of the prize went to David Baker for computational protein design.
03Has AI actually created a real drug yet?
Not an approved one. The furthest along is Insilico's rentosertib, an AI-discovered and AI-designed drug that showed positive Phase II results in a peer-reviewed journal and has entered Phase III trials. No fully AI-designed drug has FDA approval yet.
04Can AI diagnose me better than a doctor?
In controlled tests it can look that way, but those comparisons often handicap the human doctors, and medical AI still hallucinates under messy real-world input. AI is a powerful second opinion and pattern-spotter, not a replacement for a clinician who examines you and is accountable for your care.
05Is AI already used in real hospitals?
Yes. The FDA has authorized 1,451 AI-enabled medical devices, about three-quarters of them in radiology, where they help doctors read scans faster and flag findings. These tools assist physicians rather than replace them.
06Is it safe to ask ChatGPT or Claude about my health?
For understanding — explaining a diagnosis, prepping questions, decoding lab results — it can be genuinely helpful. But it should never be your final answer on a medical decision. Verify anything important with a professional, and be extra skeptical of confident or alarming answers.
The bottom line: John Jumper's move is a small news item with a big meaning. The smartest people in AI now believe the technology's most valuable work is in science and medicine, and the early results — a real drug in late-stage trials, 1,451 AI tools cleared for the clinic — suggest they are onto something. Just keep both halves of the story in view. AI is becoming a remarkable instrument for the people who cure diseases. It is not, and will not soon be, the doctor sitting across from you.
