Cancer drug development is one of the slowest and most expensive undertakings in medicine. A September 2026 review in the journal Cancers found that only about 4% of cancer drug candidates that reach human testing ever get approved. And that after roughly 14 years of work and upward of $1.2 billion per drug.

Enter artificial intelligence (AI). The evidence through 2026 shows AI is genuinely changing how cancer drugs are found, tested and delivered.

But no single AI-based “cure” to cancer is on the horizon, and one might think could occur with super-intelligence. This is partly because cancer is not just one disease. It is more than 200 diseases driven by different mutations, so there is no single cure to find.

Yet what the data show instead are incremental gains: drugs aimed at targets once thought unreachable, treatments that work better in specific patient groups, cancers caught at earlier stages, trials that enroll a year sooner. Here are seven ways AI is speeding cancer cures.

1. AI Is Helping To Aim Drugs At Targets Once Considered Undruggable

Many of the proteins that drive cancer are shaped in ways that are challenging for chemists to target. Yet with AI, researchers can now predict a protein’s three-dimensional shape and simulate how a candidate drug might lock into it. This work behind DeepMind’s AlphaFold earned Demis Hassabis and John Jumper a share of the 2024 Nobel Prize in Chemistry.

Investors are betting heavily on this technology. Isomorphic Labs, the Alphabet spinout built on AlphaFold, closed a $2.1 billion round in May, among the largest ever in AI drug discovery, and aims to start human trials by the end of 2026.

2. AI-Discovered Drugs Are Clearing Safety Trials More Often

A 2024 analysis by Boston Consulting Group researchers looked at roughly two dozen AI-derived molecules that had finished phase 1 trials. Phase 1 trials are small studies that test whether a drug is safe. The small sample of AI-derived molecules passed phase 1 between 80% and 90% of the time, compared to a historical average closer to 40% to 65.

Yet in phase 2 trials where a drug must show it actually helps patients, AI-derived molecules succeeded about 40% of the time — similar to other drugs.

Writing in ProMarket in July, Michael Santoro, a Professor Management and Entrepreneurship at the Leavey School of Business, Santa Clara University argued that AI has automated the cheap part of drug development and left the expensive part alone.

As of September 2026, no AI-designed drug had been approved. Only one, Insilico Medicine’s rentosertib, has reached a phase 3 trial — for lung fibrosis, not cancer.

3. Personalized Cancer Vaccines Depend On Computational Prediction

On Aug. 19, Merck and Moderna announced that a personalized mRNA therapy called intismeran autogene, given alongside the immunotherapy Keytruda, kept melanoma patients free of recurrence longer than Keytruda alone after surgery. It is the first late-stage success for an mRNA cancer therapy and for an individualized treatment designed for a single patient.

Here’s how it works: doctors sequence the tumor and healthy tissue, then software decides which of the tumor’s mutations will produce protein fragments the immune system is most likely to attack. Up to 34 go into each dose. That selection is the computational core. Better algorithms should mean better picks.

Five-year results from the earlier mid-stage trial, presented at ASCO in 2026, showed a 49% lower risk of recurrence or death. The program now spans nine trials, including lung, bladder and kidney cancer.

4. AI Mammography Catches Cancers Standard Screening Can Miss

Early detection remains the most reliable route to a cancer cure. and the strongest evidence for AI in cancer care comes from breast screening.

The Swedish MASAI trial randomized 105,934 women to AI-assisted mammogram reading or the European standard of two radiologists per scan. Full results in The Lancet in January 2026 showed that AI caught more cancers overall: 80.5% versus 73.8%, with no rise in false alarms. It also did not let more cancers slip through between screening rounds, the trial’s key safety test. Earlier analyses found a 29% rise in detection.

Yet, a commentary in the same journal noted that per 1,000 women screened, AI found 1.4 extra cancers but prevented only 0.2 of those that surface between screenings — suggesting the benefit may not be as impressive at the population-level.

5. AI Is Moving Some Diagnoses Earlier Through Blood Tests

Multi-cancer blood tests hunt for fragments of tumor DNA in the bloodstream, using machine learning to recognize chemical tags that signal cancer and predict which organ it came from.

Yet, the first randomized trial of one has delivered mixed results. NHS-Galleri followed about 142,000 people aged 50 to 77 in England across three annual rounds. It missed its main goal of reducing advanced cancers: a drop in stage IV cases was offset by a rise in stage III diagnoses.

The company also presented secondary findings at ASCO in May 2026 it considers more encouraging: stage IV diagnoses of 12 deadly cancers fell 14% overall, early-stage diagnoses rose 16%, and cancers first found through emergency care dropped 25%. However, the figures are sponsor-supplied and not yet peer-reviewed.

Finding fewer cancers at stage IV is a meaningful signal. Whether it means fewer deaths is still unproven .

6. A Routine Slide Can Now Predict A Tumor’s Genetics

Targeted therapy depends on knowing which mutations drive a tumor, and genetic sequencing takes weeks while consuming biopsy tissue that is often scarce.

AI models trained on huge libraries of digitized slides can now infer some of that information from inexpensive stains that nearly every biopsy already gets. The clearest example is EAGLE, built at Memorial Sloan Kettering and Mount Sinai to flag EGFR mutations in lung cancer. Results published in Nature Medicine in July 2025 showed it could cut the need for rapid genetic tests by up to 43% while preserving tissue for fuller sequencing. Positive cases still go for confirmation.

The open question is whether such tools are applicable across patient types. A 2026 JAMA Oncology study found that EAGLE held up across American and European patients. But it was less accurate in certain populations and situations, falling to about 0.68 on a zero-to-one accuracy scale in patients of Asian ancestry and 0.66 in samples from the lining of the lung.

A reliable slide-based signal could bring precision medicine to hospitals without sequencing labs. But it should only be applied in populations where the models have actually been tested.

7. AI Could Fix The Slowest Step: Trial Enrollment

Few adults with cancer ever join a clinical trial, and many studies fall behind on enrollment. Matching a patient to a trial means reading dense eligibility rules with data from complicated medical records. This is exactly what large language models are good at.

Researchers at the National Institutes of Health built TrialGPT which judges eligibility rule by rule with 87% accuracy. This is close to the 89% to 90% rate when eligibility is judged by humans. It also cut screening time by 42.6%. Importantly, the software was tested on synthetic records, so real-world performance is unproven.

And faster matching may be not faster enrollment: open slots, travel, cost and clinician judgment remain persistent bottlenecks.

A more radical use of AI is challenging the rules themselves. Trial Pathfinder, a Stanford and Genentech tool, emulated completed lung cancer trials using records from 61,094 patients and found many standard exclusions minimally affected the results.

Loosening them more than doubled the eligible pool without impacting the trial results.

Here’s What AI Can’t Speed Up

Here’s the issue: biology still has to be tested in cells, animals and people. No model shortens the years required to learn whether a treatment helps people live longer.

Regulators are also still writing the rules. The FDA issued draft guidance in January 2025, and in January 2026 it joined the European Medicines Agency in releasing ten shared principles for the field.

The realistic promise, at least in the near term, is not an super-intelligent AI that churns out cancer cures. It is a system in which fewer drugs fail late, more cancers are caught early, and trials deliver answers in months rather than years.