Late in 2026, if plans hold, a volunteer will receive a first dose of a cancer drug candidate. They will not notice how the molecule was born, and that is the point. Nobody in the room drew it by hand. Software proposed it, scored it and refined it before any glassware touched it. That moment would mark a first for AI-designed drugs.
For years, AI in medicine meant promises and paper predictions. Now the field has something better: AI-designed drugs in human trials. The shift from software demo to dosed patient is the real test of the whole approach. This post walks through the companies carrying that test. If you want the fundamentals first, read our AI in medicine cornerstone, then return here.
In this guide
- AI-designed drugs: the 30-second answer
- What you’ll learn
- What AI-designed drugs actually mean
- Isomorphic Labs: from AlphaFold to a planned cancer trial
- Insilico Medicine: the candidate already in Phase 2
- Recursion, mergers and the wider drug pipeline
- Where AI-designed drugs stand in the pipeline
- What happens next in human trials
- Keeping the hype in check
- Frequently asked questions
- The bottom line
AI-designed drugs: the 30-second answer
Here is the short version. AI-designed drugs are medicines whose targets or molecules were chosen with machine learning rather than by hand. Several AI-designed drugs now sit in clinical trials, from cancer to lung fibrosis. Isomorphic Labs, Insilico Medicine and Recursion are the names to know. Humans still run every lab and every trial. The models just aim the search at better bets first.
Key takeaways
- Isomorphic Labs plans its first human trials of AI-designed drugs around late 2026, focused on cancer.
- Insilico Medicine dosed a first patient in a Phase 2 trial of its AI-discovered fibrosis molecule in 2023.
- Recursion merged with Exscientia in late 2024 and reports around 30 percent lower research costs.
- Industry trackers count dozens of AI-discovered candidates in clinical testing worldwide, with more arriving each year.
- No approved medicine is purely AI-made yet, because human trials still take years and most candidates fail.
What you’ll learn
The route through this guide
- What the phrase AI-designed drugs really covers
- How Isomorphic Labs went from protein folding to a cancer plan
- What Insilico Medicine proved with a molecule called INS018_055
- Why Recursion and Exscientia merged their platforms
- Where clinical trials fit into the AI drug development story
- Which limits still decide what reaches patients
What AI-designed drugs actually mean
The label sounds futuristic, so it needs a careful definition. AI-designed drugs are not chemicals a computer invented alone. Researchers still choose the disease, judge the chemistry and run every experiment. What changed is where the first ideas come from.
In practice, AI touches a candidate at two points. Models can pick the biological target, the protein linked to a disease. Software can also design the molecule that binds that target. Some candidates use one kind of help, some use both.
- Target hunting: models scan biology and nominate proteins worth attacking.
- Molecule design: generative systems draft structures tuned to the chosen target.
- Trial planning: algorithms help shape studies and find suitable volunteers.
The design step is the headline act. Our guide to generative AI molecule design explains how models invent structures that have never existed. That ability is what separates this wave of drug development from older computer screening. It is also why the candidates below exist at all.
Definitions matter for another reason: accountability. If a molecule fails, blaming the algorithm helps nobody. Teams that document what the model suggested, and what humans changed, learn faster. Clear documentation turns even a failed candidate into a lesson. Paper trails of that kind also build trust with regulators.
Isomorphic Labs: from AlphaFold to a planned cancer trial
Start with the company carrying the biggest spotlight in AI-designed drugs. Isomorphic Labs is an Alphabet company spun out of Google DeepMind to design medicines with AI. It inherited the protein folding work behind AlphaFold. In May 2026, it raised a reported 2.1 billion dollars, its first major outside raise.
Money on that scale signals intent. The company says its first human trials of AI-designed drugs are slated for around the end of 2026. That plan focuses on cancer, according to press coverage. Timelines in this industry can slip, so treat the date as a plan rather than a promise. Even so, the field has never been this close.
Why does one funding round matter so much? Isomorphic Labs had been funded by Alphabet until this raise, so outside money is new. The raise also pays for lab capacity and trial preparations at once. Meanwhile, competitors watch the timeline closely, because a first dosing would reset expectations for the whole field.
AlphaFold explains part of the confidence. The protein prediction work behind the company reshaped a slow corner of structural biology. Its stated focus is designing medicines with AI from the start. Press reports describe the team as one of the most closely watched in the field.
Scale is the quiet question behind the funding. Designing a molecule is cheap compared with testing it. Human trials cost more than any algorithm, and they cannot be compressed by enthusiasm. Investors know this, which is why the trial date matters more than the raise. That is also why labs, hospitals and regulators stay central to the plan.
Insilico Medicine: the candidate already in Phase 2
One candidate has already gone further than a plan. Insilico Medicine dosed the first patient in a Phase 2 trial of INS018_055 in 2023. The molecule targets idiopathic pulmonary fibrosis, a scarring lung condition with few good options. An AI platform helped pick the target, and another helped design the molecule.
Why does fibrosis make a good test case? Idiopathic pulmonary fibrosis scars the lungs over years, and treatment options remain limited. A genuinely new chemical type would matter to patients waiting for progress. It would also validate the platforms that built the molecule.
Together, those two steps show a full pipeline at work. Target selection and molecule design are the two hardest early steps in drug development. Insilico reports that its platforms handled both, with chemists steering every round. Peer review of the full story is still catching up, so hold the claims lightly.
Progress of this kind changes the argument. Critics long said AI could only impress in simulations. A molecule discovered with AI, dosed in people, is evidence of a different grade. Other companies noticed, and the race for human data began in earnest.
Numbers stay scarce here, and that is worth stating plainly. Companies report milestones, not full datasets, so independent verification takes time. Journals and regulators see more than press releases do. Until they publish, treat every platform claim as promising but unfinished.
Recursion, mergers and the wider drug pipeline
Isomorphic and Insilico are not alone. Recursion, an AI biotech, merged with fellow AI drug developer Exscientia in late 2024. Consolidation of this kind usually means the technology has moved past the demo stage. Exscientia and Recursion each brought their own platforms to the table. The combined company reports its platform cuts research costs by around 30 percent.
Treat that number as a company claim, not an independent audit. Still, the direction is consistent across the industry. Meanwhile, the software itself is getting more capable. Systems now plan experiments and adjust course, a shift our guide to what AI agents can do explains.
Beyond the big names, the bench is crowded. Industry trackers count dozens of AI-discovered candidates in clinical testing worldwide, with more entering each year. Most come from partnerships between AI firms and established pharmaceutical companies. The drug pipeline now has machine help at nearly every stage.
Money follows capability, and so do partnerships. Industry trackers follow the deals as closely as the science. Each partnership moves more machine-picked candidates toward the clinic. As a result, bench chemistry now shares space with code review.
What does this mean for job seekers and watchers? Platform companies hire biologists, chemists and data scientists together. The merger era rewards people who speak both languages. Much of the underlying research still comes from university labs.
Where AI-designed drugs stand in the pipeline
A snapshot helps here. The table below maps the journey of AI-designed drugs, from first target pick to possible approval. Note how the early stages are crowded with results, while the last stage waits on time.
The journey of an AI-designed drug, stage by stage
| Stage | What the machines do | Where it stands today |
|---|---|---|
| Pick a target | Models rank disease-linked proteins and flag the most promising ones | Standard practice across AI-first labs |
| Design a molecule | Generative models propose structures tuned to the chosen target | INS018_055 reached Phase 2 in 2023 |
| Test in the lab | Robotics runs the experiments while models refine the next round | Core workflow at Recursion and peers |
| Dose human volunteers | People receive the candidate while clinicians watch safety closely | Underway, with Isomorphic trials planned around late 2026 |
| Win approval | Regulators review years of evidence before any launch | Still pending for every purely AI-made drug |
Read the table from the bottom up and you see the honest picture. Approval is the row with no ticks yet. Nothing approved today is purely AI-made, and trials still take years. As in any pipeline, most candidates fail along the way.
Two lessons sit inside that table. Early stages got cheap because data went digital first. Late stages stayed expensive because bodies, regulators and biology refuse to move at computer speed. Closing that gap is the industry’s next problem.
Tracking the field is easier than it used to be. Public trial registries list candidates and their sponsors openly. Industry databases flag which studies involve AI-discovered molecules. Curious readers can verify claims without trusting any press release, including ours.
What happens next in human trials
Dosing a first patient is a beginning, not a finish line. Human trials run in phases, each asking a harder question. Early phases check safety in small groups. Later phases test whether the drug actually helps, in hundreds of people. Each phase gates the next, and regulators hold the keys.
AI helps on this stretch of the road too. Our article on AI in clinical trials shows how models match patients to studies and speed recruitment. Faster matching matters, because slow enrollment delays every medicine waiting behind it. For a volunteer in a fibrosis or cancer trial, the science stays identical.
Timelines deserve realism here. A Phase 2 trial that dosed its first patient in 2023 moves slowly, by design. An Isomorphic cancer trial, even if it starts on schedule, would need years before any approval claim. Press reports describe the end of 2026 as a start, not a finish.
What should a curious reader watch for? Safety readouts come first, then efficacy signals, then larger confirmatory studies. Each step filters candidates, and most will not survive them. That is normal, and it protects the people who volunteer.
Patience is the hardest ingredient. A trial planned for late 2026 will report meaningful results years later, not months. Anyone following along should expect quiet stretches between headlines. Still, quiet stretches are when the real work happens.
Keeping the hype in check
Every wave of medical technology attracts overpromising, and this one is no exception. Headlines compress years of work into single words like breakthrough. The reality is steadier and less cinematic. In truth, progress arrives one verified result at a time. AI-designed drugs remain candidates until regulators say otherwise.
Skepticism should stay proportionate, though. The candidates are real, the money is real and the timelines are public. What has not changed is biology’s habit of surprising late-stage testing. Nature does not read press releases, and neither do regulators. Our honest guide to what AI in medicine still cannot do covers those boundaries in detail.
Overall, the smart position sits between the extremes. This is genuine progress, visible in trials rather than slideware. It is also early, and history says most candidates will not make it. Patients benefit either way, because failure data teaches the next model.
A simple habit helps when reading this news. Check whether a claim says discovered, designed or approved. Those words sit years apart in a drug pipeline. Most stories in 2026 describe the first two, and that is still news worth having.
Perspective helps, too. Software did not replace the scientific method; it joined it. Hypotheses still need experiments, and experiments still need honest writeups. The tools changed the tempo, not the rules.
Trial watchers should track three signals from here. First, the actual trial start dates, since plans slip. Second, safety readouts, because a first in human study lives or dies there. Third, whether regulators publish guidance for AI discovered molecules. Watch those three, and the noise will sort itself out. The scoreboard, not the press release, decides the verdict.
Frequently asked questions
What are AI-designed drugs?
They are medicines whose biological target, molecule or both were chosen with machine learning. Scientists still design the studies, run the labs and make the final calls. The label describes smarter search, not an automated drug factory. In short, the model proposes and the chemist decides.
Which companies have AI-designed drugs in or near human trials?
Isomorphic Labs plans first human trials around late 2026, focused on cancer. Insilico Medicine dosed a first patient in a Phase 2 fibrosis trial back in 2023. Recursion runs a broad pipeline built on its merged platform, with more candidates behind them. Dozens more candidates from AI pipelines are in earlier testing stages.
When will the first approval happen?
Nobody can say yet, and honest researchers avoid promising dates. Human trials take years, and most candidates fail at some stage. An AI-designed drug will clear the bar only when the evidence is complete. Watch trial milestones instead of headline dates.
Is Isomorphic Labs part of Google?
It is an Alphabet company, spun out of Google DeepMind to design medicines with AI. Alphabet is Google’s parent firm, so the link is real but not simple. The company also raised outside funding in 2026, which added new investors. Its research direction stays focused on AI-designed medicines.
Are AI-designed drugs safe?
They follow the same trial rules as any other candidate. Regulators review safety data before and during every human study. Volunteers in early trials are watched closely, and they can leave a study at any time. In fact, no candidate skips these checks because software helped design it.
How much money is behind this wave?
The clearest signal is Isomorphic Labs’ reported 2.1 billion dollar raise in May 2026. It was the company’s first major outside funding round. Across the field, trackers count dozens of candidates in clinical testing worldwide. In addition, most large pharmaceutical firms now run AI drug development teams of their own.
The bottom line
The first AI-designed drugs are no longer a pitch, they are a calendar entry. Isomorphic Labs plans cancer trials for around late 2026. Insilico’s fibrosis molecule has been dosing patients since 2023. Recursion and dozens of smaller efforts keep widening the drug pipeline. Behind each line sits years of quiet model building.
For patients, the change is real but slow. Meanwhile, the industry finally has a scoreboard built from human data. Watch the trial dates, read past the press releases, and you will know exactly how the story unfolds. Wherever it lands, the era of machine-designed medicine has clearly started.
Sources
- Isomorphic Labs company information and funding coverage, 2025-2026 – isomorphiclabs.com and press reports
- Insilico Medicine pipeline updates on INS018_055 first patient dosing, 2023 – insilicomedicine.com
- Recursion reports on the Exscientia merger and platform cost claims, 2024-2026 – company reports
- Google DeepMind background on AlphaFold and the Isomorphic Labs spinout – deepmind.google
- Industry pipeline trackers of AI-discovered candidates in clinical testing – field databases
- Press coverage of AI drug development timelines, 2025-2026 – science and trade press