An AI-designed antibody just reached Phase III trials, and two more labs are racing to follow
Generate Biomedicines now has an AI-designed antibody running in multiple Phase III trials for asthma, partnered with Amgen and Novartis. That single fact matters more than any model benchmark published this year, because Phase III is the stage where a drug candidate has already survived Phase I safety testing and Phase II efficacy signals. An AI system did not just fold a protein in a demo; it produced a molecule good enough to clear two separate human-trial gates and attract two Big Pharma co-development deals.
For three years, the pitch from AI-drug-design startups has been consistent: generative models can search chemical and biological space faster than wet-lab screening, so the industry should expect AI-originated drugs any year now. 2026 is the year that stopped being a promise and became a pipeline with actual trial-registry entries. Generate's lead candidate targets TSLP, a cytokine implicated in severe asthma, and the fact that it reached Phase III under two different pharma partnerships is a stronger signal than a single company's internal claim.
Who else is actually in the clinic, not just fundraising
Isomorphic Labs, DeepMind's drug-design spinout, has been preparing to dose its first patients for roughly four years since its 2021 founding. The company raised $600 million in March 2025 led by Thrive Capital and recently introduced IsoDDE, a proprietary drug design engine the company says outperforms AlphaFold 3 on accuracy for its internal benchmarks. Isomorphic's first candidates target oncology indications. As of this writing, the company has described itself as preparing to enter clinical trials rather than confirming dosed patients, which is a meaningfully earlier stage than Generate's Phase III programs.
Xaira Therapeutics launched in 2024 with over $1 billion in funding, the largest biotech launch round in years, built around an antibody-design platform called X-Design. Xaira's CEO has said two candidates, internally labeled XA-1 and XA-4, are planned to advance into human studies, but the company has not disclosed a timeline. That reticence is itself informative: Xaira has had two years and a billion dollars, and the public story is still "planned," not "dosed" or "Phase III."
Why the gap between these three companies matters
The difference between Generate, Isomorphic, and Xaira is not model quality in the abstract — it is how each company chose to prove the model works. Generate paired its AI platform with two of the largest pharma companies in the world and let them run the trials under their own regulatory infrastructure. That de-risks the clinical execution problem, which is usually the part that kills biotech startups, not the molecule-design problem. Isomorphic and Xaira are still building out their own clinical and regulatory muscle, which is slower but keeps more economic upside in-house if a candidate succeeds.
This is the actionable lesson for anyone evaluating AI-biotech claims: ask whether a company's AI-designed candidate is running inside an established pharma company's trial infrastructure or inside the startup's own nascent one. A press release about a "breakthrough AI-designed molecule" entering Phase I means something very different depending on who is running the trial and who bears the regulatory risk. Phase III, specifically, is the number to watch — it is the stage immediately before an FDA submission, and getting there means a molecule has already shown it does not kill patients and does show a treatment effect.
What to actually track going forward
Three concrete markers will tell you whether AI drug design keeps advancing or stalls out: first, whether Generate's TSLP antibody reads out positive Phase III data, since a flop would be the first major reality check for an AI-originated molecule at this stage; second, whether Isomorphic actually dosed a first patient, since the company has described itself as "preparing" for over a year; third, whether Xaira discloses a trial start date for XA-1 or XA-4, since silence after a $1 billion raise and two years of operation is a signal in itself. None of these require insider access — ClinicalTrials.gov registrations and quarterly pharma-partner disclosures will show the answer before any company's own press release does.