Do AI-discovered cancer drugs work better in clinical trials?
AI drug discovery has produced a growing pipeline of cancer drug candidates, and early data suggest AI-discovered molecules clear phase 1 safety trials more often, but phase 2 success looks similar to conventional drugs. A peer-reviewed 2024 analysis reported 80% to 90% phase 1 success and about 40% in phase 2 on a small sample.
Developing a cancer drug conventionally takes about 14 years and about 1.2 billion US dollars, and most candidates fail, according to a 2026 review. AI promises to shorten the early steps: choosing targets, designing molecules and predicting toxicity. For patients, the question is simpler: does it lead to treatments that work, and can they access them?
This page summarises what peer-reviewed analyses and company disclosures show, which AI-designed cancer drugs are in human trials, what is still unproven, and how patients, including those in India and China, can approach clinical trials of new agents.
Key facts on AI drug discovery in oncology trials
Peer-reviewed figures and company-reported trial data, labelled by source:
- Peer-reviewed analysis (Drug Discovery Today, 2024): phase 1 success of 80% to 90% for AI-discovered molecules versus historic averages of roughly 40% to 65%; oncology made up about half of AI-discovered molecules in phases 1 and 2
- Phase 2 in the same analysis: ten AI-discovered molecules had completed trials and four succeeded, a limited sample according to the authors
- A 2026 review states that phase 2 results remain consistent with historical averages of about 40%
- Conventional oncology development: about 14 years and about 1.2 billion US dollars per drug, with low trial success rates (2026 review)
- REC-617 (CDK7 inhibitor, Recursion): interim phase 1 data in December 2024 reported mostly grade 1 to 2 adverse events, no maximum tolerated dose reached and a confirmed partial response in one patient with platinum-resistant ovarian cancer (company-reported)
- REC-1245 (RBM39 degrader, Recursion): preliminary data in 16 patients, with any treatment-related adverse event in 62.5% and grade 3 in 6.2% (company-reported, May 2026)
- ISM3412 (MAT2A inhibitor, Insilico Medicine): first patient dosed in June 2025 in a global phase 1 trial (company-reported)
Key takeaways: AI drug discovery and cancer clinical trials
- AI-discovered molecules showed high phase 1 success (80% to 90% in a 2024 analysis), mainly reflecting safety and tolerability.
- Phase 2, which tests whether a drug works, shows success near historical averages of about 40%, on a limited sample.
- Named oncology candidates, including REC-617, REC-1245 and ISM3412, are in early trials, and their efficacy is not yet established.
- We did not find a cancer drug approved on the basis of AI-driven discovery in the sources reviewed.
- Patients can reach new agents mainly through early-phase clinical trials, where strict eligibility and safety monitoring apply.
How is AI used in cancer drug discovery?
AI supports several steps. Target discovery uses large biological datasets to propose proteins or pathways that drive a cancer. Molecule design uses generative models to propose compounds with the desired shape and properties, predicting binding and toxicity before synthesis. Biomarker and patient-selection models aim to identify who is likely to respond. In trials, AI helps with eligibility screening and outcome prediction.
Recursion states that REC-617 was designed with its AI-led platform, with only 136 novel molecules synthesised from hit to candidate in less than 12 months. Insilico states that ISM3412 was designed with its generative chemistry engine and nominated as a preclinical candidate in May 2022. These speed claims come from the companies and are not independent comparisons.
What do peer-reviewed analyses say about AI-discovered drug success rates?
The most cited analysis, by Boston Consulting Group authors in Drug Discovery Today in 2024, found phase 1 success of 80% to 90% for AI-discovered molecules, substantially above historic industry averages of about 40% to 65%. In phase 2, ten AI-discovered molecules had completed trials and four were successful, a rate near 40% that is in line with historical averages and rests on a small sample. Oncology accounted for about half of the AI-discovered molecules in phases 1 and 2.
Later publications repeat these figures with caveats. A 2025 editorial on AI in cancer pharmacology notes that the 40% phase 2 estimate is based on a small sample and is not specific to oncology, and that while some candidates remain in development, others have failed in trials.
Key success-rate figures for AI-discovered drugs
From the 2024 analysis and later reviews:
- Phase 1 success: 80% to 90% for AI-discovered molecules versus about 40% to 65% historically
- Phase 2: 4 of 10 completed AI-discovered molecules were successful (about 40%)
- Oncology share of AI-discovered molecules in phases 1 and 2: about 50%
- Interpretation: a strong early safety signal, but an efficacy signal not yet distinguishable from conventional development
Which AI-designed cancer drugs are in human trials?
Several oncology candidates are in early trials. Recursion reports REC-617, an oral CDK7 inhibitor, in the ELUCIDATE phase 1/2 study in advanced solid tumours; REC-1245, an RBM39 degrader, in the DAHLIA phase 1/2 study; and REC-4539, an LSD1 inhibitor for solid tumours including small cell lung cancer and AML, with the first patient dosed in April 2026. Insilico reports that ISM3412, a MAT2A inhibitor, entered a global phase 1 trial in locally advanced or metastatic solid tumours in June 2025.
Recursion also reports that an IND application has been cleared for REC-7735, a PI3Kα H1047R-selective inhibitor, with a phase 1/2 study planned for the second half of 2026 in PIK3CA H1047R-mutant solid tumours. These are early-phase studies: their main goals are safety, dose and early signals of activity.
What do early data for AI-designed oncology drugs actually show?
Early data are encouraging on tolerability but thin on efficacy. For REC-617, Recursion reported mostly grade 1 to 2, on-target and reversible adverse events, no maximum tolerated dose reached and no discontinuations because of adverse events in the interim analysis. Efficacy was not an endpoint, and a confirmed durable partial response by RECIST was reported in one patient with platinum-resistant ovarian cancer. For REC-1245, preliminary data in 16 patients reported treatment-related adverse events in 62.5%, with 6.2% grade 3.
These are company announcements of interim phase 1 data. One patient responding is a signal worth following, not evidence that a drug works. Peer-reviewed publication and larger phase 2 trials are needed before any conclusion about benefit.
Why do AI-discovered drugs still fail in clinical trials?
Designing a molecule that binds a target is only part of the challenge. Many oncology failures arise because the target biology does not hold in patients, because tumours develop resistance, because the right patients are not selected by biomarker, or because toxicity limits the dose that can be given. AI can optimise molecules but cannot show that a target matters in humans. The 2026 review notes that oncology trial success remains low, and the editorial notes that some AI-designed candidates have failed in trials.
Matching the right drug to the right patient also depends on biomarker testing and trial design, which is where AI-assisted selection and clinical trial matching may help. Still, early-phase results do not tell patients whether to expect benefit.
Limits of the AI drug discovery evidence
Headlines about AI-designed drugs often outpace the data.
- The phase 2 success estimate rests on a small sample and is not specific to oncology
- Most oncology data are interim phase 1 results announced by the companies, not peer-reviewed publications
- A response in a single patient cannot show efficacy
- We did not find a cancer drug approved on the basis of AI-driven discovery in the sources reviewed, and approval status can change
Who could ask about clinical trials of AI-designed cancer drugs?
Early-phase trials are options for some patients after standard therapies, not a default first choice. Situations to discuss with an oncologist:
- Your cancer has progressed after standard treatments and few options remain
- Your tumour carries a specific alteration, such as PIK3CA H1047R, that a trial targets
- You are fit enough, with adequate organ function, to meet early-phase eligibility criteria
- You can travel to a trial site and attend frequent visits, biopsies and scans
- You understand that phase 1 trials mainly test safety and dose, and that benefit is uncertain
- You want a specialist to check your records against open early-phase trials
How do patients access early-phase cancer trials of new agents?
Search ClinicalTrials.gov and national registries by cancer type, alteration and phase, then read the eligibility criteria and site lists. Early-phase trials may be open at only a few centres, may have limited slots and typically require recent scans, biomarker results and adequate organ function. Sponsors and investigators decide eligibility, and participation is voluntary with informed consent.
Trial drugs are usually provided by the sponsor, but costs such as travel, accommodation and routine care may fall on the patient depending on the trial and country, so ask the trial team what is covered.
How does AI drug discovery compare with conventional drug development?
Conventional oncology development is long and costly, and success is low. AI aims to compress early discovery: Recursion states that REC-7735 reached development candidate in 10 months and 242 compounds from the first novel hit, and the REC-617 programme synthesised 136 novel molecules from hit to candidate in under 12 months. These speed claims are company-reported and have not been independently compared with conventional timelines.
Once a molecule enters humans, it faces the same trial path as any drug: safety, dose finding, efficacy and regulatory review. The phase 1 data suggest AI may help with molecular properties that influence tolerability, but phase 2 results so far are not distinguishable from those of conventional drugs.
AI drug discovery evidence and access by region
What has been reported, with gaps stated:
- Global counts: numbers of AI-enabled assets in human trials differ by source; one 2026 ASCO abstract, as summarised by an industry tracker, counted 117 assets across 63 companies
- Named oncology candidates: REC-617, REC-1245, REC-4539 and ISM3412 are in phase 1 or phase 1/2 trials announced by Recursion and Insilico
- China and India: Insilico describes a global multicentre trial for ISM3412, but we have not verified open sites in either country
- Approvals: no approval of an AI-discovered cancer drug was identified in the sources reviewed
Can patients in India or China join trials of AI-designed cancer drugs?
Insilico describes the ISM3412 phase 1 study as a global multicentre trial, but we have not verified which countries or sites are recruiting, and we have not verified open sites in India or China for any AI-designed cancer drug named on this page. Recruitment status changes often and should be checked on the registry and with the sponsor.
What happens next for AI drug discovery in oncology?
Several readouts are scheduled. Recursion expects further REC-1245 phase 1 data in the second half of 2026 and early combination data for REC-617 in the first half of 2027, while REC-7735 is expected to start its phase 1/2 study in the second half of 2026. These are company expectations and can change.
The decisive evidence will be randomised phase 2 and phase 3 results showing that AI-designed drugs extend life or improve quality of life compared with standard treatment, followed by regulators’ decisions on approval. Until then, the most credible use of AI in drug development is as a tool that speeds early stages, with clinical trials remaining the test of benefit.
Questions to ask your oncologist about trials of new AI-designed drugs
- Is there an early-phase trial that matches my tumour type and genetic changes?
- What is known about the drug’s safety, and what are the main risks?
- What would joining involve: visits, biopsies, scans, travel and time off?
- What happens to my standard treatment options if I join a phase 1 trial?
- Who confirms my eligibility, and what records will they need?
- Can a trial specialist review my case before I decide?
Sources
Where the figures in this guide come from:
PMC: Editorial, AI research in cancer pharmacology (2025)
Recursion Pharmaceuticals: first quarter 2026 financial results and business update (May 6, 2026)
BioSpace: Recursion reports second quarter 2026 financial results (August 5, 2026)
IntuitionLabs: AI-discovered drugs in clinical trials 2026, pipeline tracker (secondary source)