10 AI Tools That Promise to Cure Disease, and Where Each Drug Pipeline

How these ten AI drug and diagnostic tools were ranked

Every figure in the ranking below comes from Google Trends worldwide web search data covering the 12 months to 8 September 2026. Google Trends reports a relative index between 0 and 100 rather than an absolute number of searches, so a Google Trends index says how much attention one search term received compared with other search terms, and a Google Trends index never says how many people searched.

Four separate Google Trends comparisons were run, and the search term "alphafold" appeared in all four comparisons as a shared anchor, which allows the ten tools to be placed on a single scale. The rescaling arithmetic across the four comparisons is this post's own calculation and is not something Google Trends itself reports.

Two of the ten search terms carry known contamination that inflates the measured index. The search term "alphafold" also captures searches for a luxury mobile phone marketed as the Vertu AlphaFold, because Google Trends lists "vertu alphafold", "vertu phone" and "vertu alphafold price in india" among the rising related queries for "alphafold". The search term "tempus ai" also captures share-price searches, because Google Trends lists stock-related queries such as "spacex stock" and "cerebras stock" among the rising related queries for "tempus ai".

AlphaFold and Isomorphic Labs: protein structure prediction aimed at drug design

AlphaFold is a protein structure prediction system from Google DeepMind, and AlphaFold predicts the folded three-dimensional shape of a protein from the protein's amino acid sequence. AlphaFold does not by itself design a medicine, and AlphaFold is best understood as an input into drug design rather than as a treatment.

Isomorphic Labs is the drug design company inside the Alphabet group, and the Isomorphic Labs website states an ambition "to one day solve all disease with the help of AI". The same Isomorphic Labs website describes the Isomorphic Labs Drug Design Engine as unlocking "a new frontier beyond AlphaFold", and the same website discloses no named drug candidate and no clinical trial status.

Clinical Trials Arena reported on 7 July 2025 that Isomorphic Labs president Colin Murdoch said the company was "preparing to dose the first patients in clinical trials", with oncology named as the first therapy area. The claim that Isomorphic Labs is preparing to dose patients comes from Clinical Trials Arena quoting a company executive, and the claim was not accompanied by a candidate name, a trial registration number or a start date.

Insilico Medicine: an AI-discovered lung fibrosis drug with published Phase 2a data

Insilico Medicine used generative chemistry to design rentosertib, formerly named ISM001-055, which inhibits the enzyme TNIK, a protein kinase that Insilico Medicine also identified as a target for idiopathic pulmonary fibrosis. Idiopathic pulmonary fibrosis is a progressive scarring disease of the lung for which no existing therapy reverses the disease course.

According to PubMed, the rentosertib Phase 2a trial was published in Nature Medicine in 2025 and randomised 71 patients with idiopathic pulmonary fibrosis into four arms for 12 weeks, with 18 patients on 30 mg once daily, 18 patients on 30 mg twice daily, 18 patients on 60 mg once daily and 17 patients on placebo (DOI 10.1038/s41591-025-03743-2, PMID 40461817). The primary endpoint of the rentosertib Phase 2a trial was the percentage of patients with at least one treatment-emergent adverse event, and the percentage was similar across all four arms at 72.2%, 83.3%, 83.3% and 70.6% respectively.

The forced vital capacity result, which measures the volume of air a patient can forcibly exhale, was a secondary endpoint of the rentosertib Phase 2a trial rather than the primary endpoint. Patients on 60 mg of rentosertib once daily showed a mean forced vital capacity change of plus 98.4 millilitres with a 95% confidence interval of 10.9 to 185.9, against minus 20.3 millilitres with a 95% confidence interval of minus 116.1 to 75.6 for placebo (PMID 40461817).

According to PubMed, a 2026 network meta-analysis in BMC Pulmonary Medicine covering 35 reports and 8,983 participants placed rentosertib among the three treatments with the highest statistical probability of preserving forced vital capacity, at a standardised mean difference of 0.63 with a 95% confidence interval of 0.04 to 1.22 (DOI 10.1186/s12890-026-04342-8, PMID 42115887). The same network meta-analysis rated evidence certainty across the idiopathic pulmonary fibrosis treatment comparisons as mostly low to very low, and the same network meta-analysis found that no regimen significantly reduced all-cause mortality.

Insilico Medicine announced on the Insilico Medicine website that a Phase 3 trial of rentosertib in idiopathic pulmonary fibrosis has been initiated, and Insilico Medicine separately announced that the pan-TEAD inhibitor ISM6331 received United States Food and Drug Administration Fast Track designation in advanced mesothelioma. Fast Track designation is a regulatory process designation that can speed review, and Fast Track designation is not evidence that a drug works.

Recursion Pharmaceuticals: cell-image screening with Phase 1b/2 polyp data

Recursion Pharmaceuticals photographs treated cells at very large scale and uses machine learning to spot which compounds shift a diseased cell back toward a healthy appearance. Recursion Pharmaceuticals merged with the AI drug design company Exscientia, and the merged pipeline carries programmes originally started under both company names.

Recursion Pharmaceuticals reported on 8 December 2025 that the TUPELO Phase 1b/2 trial of REC-4881 in familial adenomatous polyposis produced a 43% median reduction in total polyp burden among 12 patients evaluable for efficacy at week 13, and a 53% median reduction at week 25 after 12 weeks off treatment, with 9 of 11 patients responding at week 25. Familial adenomatous polyposis is an inherited condition in which large numbers of polyps form in the bowel and carry a high risk of becoming cancer.

The same Recursion Pharmaceuticals release reported that 94.7% of the 19 patients across the combined Phase 1b/2 cohorts had at least one treatment-related adverse event, that Grade 3 treatment-related adverse events occurred in 15.8% of patients, and that no Grade 4 or higher treatment-related adverse events had been reported. All of the REC-4881 efficacy and safety figures in this paragraph come from the Recursion Pharmaceuticals investor release rather than from a peer-reviewed publication.

Tempus AI: molecular and clinical data used to choose cancer treatment

Tempus AI does not design drug molecules, and Tempus AI instead sells genomic testing and data services that inform which existing treatment a patient receives. The Tempus website describes the company as having "built the world's largest library of clinical & molecular data and an operating system to make that data accessible and useful, starting with cancer".

Tempus AI announced on 29 May 2026 that the United States Food and Drug Administration approved the tumour-only version of the Tempus xT CDx assay, which is a companion diagnostic test that reads tumour DNA to match a patient with a targeted therapy. A companion diagnostic approval is a regulatory clearance for a test, and a companion diagnostic approval is not the approval of a new medicine.

The Tempus AI search index of 30 recorded in this post is inflated by share-price searches, because Google Trends lists stock-related queries including "spacex stock", "sandisk stock" and "cerebras stock" among the rising related queries for the search term "tempus ai".

Generate:Biomedicines and Absci: AI-designed antibodies now dosed in humans

Generate:Biomedicines announced on 1 December 2025 that the company will start two global Phase 3 studies, named SOLAIRIA-1 and SOLAIRIA-2, of GB-0895 in severe asthma. GB-0895 is described by Generate:Biomedicines as an investigational long-acting monoclonal antibody engineered with artificial intelligence to target thymic stromal lymphopoietin, which is a signalling protein released by airway cells that drives allergic inflammation.

The same Generate:Biomedicines announcement stated that the two SOLAIRIA studies will enrol approximately 1,600 patients across more than 40 countries, that the primary endpoint is reduction in asthma exacerbations over 52 weeks, and that the Phase 3 dose is 300 mg given under the skin every six months. The same announcement stated that the earlier Phase 1 study of GB-0895 involved 96 participants with mild to moderate asthma and showed a half-life of approximately 89 days, and a half-life is the time taken for half of a dose to be cleared from the body.

Absci announced in May 2025 that the first participants had been dosed in a Phase 1 trial of ABS-101, which Absci describes as a potential best-in-class antibody against TL1A for inflammatory bowel disease. TL1A is an immune signalling protein that several companies are targeting in inflammatory bowel disease, so the ABS-101 programme is a contest against competing antibodies rather than a first attempt at an untried target.

XtalPi, Iambic Therapeutics and BenevolentAI: three smaller AI drug pipelines

XtalPi combines computational chemistry with robotic laboratories and mainly discovers molecules for partner companies rather than for the XtalPi pipeline. XtalPi announced that the PRMT5 inhibitor PEP08, developed with PharmaEngine, received clinical trial approvals, and the XtalPi model means that XtalPi progress is usually reported through a partner company's programme.

Iambic Therapeutics reported on 20 October 2025, at the 2025 ESMO Congress, that the Phase 1/1b study of the HER2 inhibitor IAM1363 produced partial responses in 28% of 18 heavily pretreated patients with measurable systemic disease and in 33% of 3 patients with measurable tumours inside the brain, at doses of 960 mg once daily and above. HER2 is a growth-signalling receptor that is altered in several cancers including breast cancer and gastric cancer, and the number of patients in the intracranial group was three, so the 33% figure rests on a single responding patient.

Iambic Therapeutics stated in the same ESMO announcement that IAM1363 moved "from program start to clinical trial initiation in just two years", and the two-year timeline is the Iambic Therapeutics claim about speed rather than about efficacy. BenevolentAI is the cautionary entry among the ten tools, because the BenevolentAI lead candidate BEN-2293 met its safety endpoint but missed both secondary efficacy endpoints in atopic dermatitis, and BenevolentAI announced on 6 February 2025 a proposed merger into Osaka Holdings that removed BenevolentAI from Euronext Amsterdam on 13 March 2025.

What none of the ten AI medicine tools has delivered yet

No drug discovered or designed by artificial intelligence had received full regulatory approval anywhere as of 8 September 2026. The most advanced published efficacy signal among the ten tools is the rentosertib forced vital capacity result, and the rentosertib forced vital capacity result came from a secondary endpoint in a 71-patient, 12-week trial (PMID 40461817).

Several artificial intelligence drug programmes have already been discontinued after entering human trials. DSP-1181, developed by Exscientia with Sumitomo Pharma for obsessive-compulsive disorder and widely described as the first AI-designed drug to enter human trials, was abandoned in 2022. BEN-2293 from BenevolentAI missed both secondary efficacy endpoints in atopic dermatitis, and VRG50635 from Verge Genomics missed its primary efficacy endpoint in amyotrophic lateral sclerosis.

The gap between a promise and an approved medicine is measured in endpoints, sample sizes and confidence intervals, and none of those three quantities is improved by the fact that a molecule was designed by a model. A reader assessing any of the ten tools should ask which endpoint was primary, how many patients were randomised, and whether the result has been published in a peer-reviewed journal rather than only in an investor release.

Frequently asked questions