The most useful AI story in oncology this quarter is a description of how drug discovery might work, not a product launch. Puja Sapra, writing in MIT Technology Review, lays out a "lab of the future" built from three parts: "Where a self-driving car uses sensors and models to navigate its environment, this system uses AI to make predictions, robotic systems to execute experiments, and instruments to generate data."1 It arrives while the clinical calendar is unusually full, so it is worth asking how much of the AI pitch can be checked against the rest of the field.
What the AI claims actually say
The piece is about AstraZeneca's approach to designing biologics, which are large-molecule drugs. Its central argument is about data, not model architecture: "Data is our differentiator. We've built an intentionally diverse portfolio across multiple disease areas and drug types. All of that data empowers us to fine-tune frontier AI models with richer, more representative training sets."1 For practitioners, that is a claim that proprietary, multimodal training data (structures, binding measurements, safety profiles, manufacturing outcomes) matters more than access to a frontier model.
The article also describes a shift toward systems that generate a candidate molecule and predict its behavior together. In Sapra's words, "The complexity of the biology goes hand-in-hand with the design of the molecule."1 For multi-specific biologics, she says models "could help identify which two or three targets to prioritize based on the underlying biology, then optimize across multiple parameters to balance a molecule's potency, stability, manufacturability, and safety."1
The piece is candid about the hardest part: "One of the hardest problems in de novo design is predicting whether a computationally generated molecule will be safe in the human body."1 Her proposed answer is virtual clinical trials using advanced cell systems and micro-scale organ models. She says these "have the potential to generate enhanced biological signals without traditional testing bottlenecks, and they're a critical missing piece in closing the loop between AI-generated designs and clinical-ready candidates."1 On throughput, she expects automation to "generate AI-ready data at a scale that traditional workflows cannot match."1
The human role is stated plainly: "Scientists will work hand-in-hand with these model systems. There will be a world where models will design molecules, then scientists will work with the systems to test those molecules and put all that data together."1 She adds that "human oversight remains at the heart of this approach—ensuring explainable and ethical AI for the benefit of patients."1
How much weight the source can bear
Our verification layer flags a problem. Of 11 claims checked from this MIT Technology Review document, none held up against our source-of-truth checks.1 We are not saying the statements are false. Most are forward-looking descriptions that are hard to check, and the piece is undated in our records even though its listed source date is 2026-07-23. It does mean that every AI claim above is the speaker's own account, with no independent confirmation in our data. Notably, the dossier contains no trial result, timeline or cost figure showing that an AI-designed oncology molecule has reached the clinic.
That gap matters because the piece's own premise is that safety prediction is the bottleneck. The loop it describes, from AI design to a clinical-ready candidate, is presented as something still being closed.1
The clinical calendar the AI has to compete with
Traditional development, by contrast, is producing checkable events. The FDA approved Tecentriq and Tecentriq Hybreza with a fluoropyrimidine and oxaliplatin for adjuvant stage III dMMR colon cancer on 2026-10-09, based on the Phase III ATOMIC study, its twelfth U.S. indication.2 Summit Therapeutics' Biologics License Application for ivonescimab, based on HARMONi trial results, carries a PDUFA goal date of 2026-11-14.3 Moderna and Merck plan to present detailed intismeran results at ESMO on 2026-10-24 and to engage regulators about potential filings.4
Real-world data is also arriving. Kyowa Kirin's final MINT analysis of 109 patients across 18 German centres reported a median time-to-next-treatment of 14.9 months, a 62.4% best overall response rate and 20.2-month median progression-free survival for mogamulizumab.5 In Europe, the CHMP recommended Gilead's Trodelvy plus Keytruda in PD-(L)1-positive first-line metastatic triple-negative breast cancer on 2026-07-24.6 These are conventional drugs with conventional evidence. They set the bar an AI-designed molecule must eventually clear.
Talent is moving toward the personalized-vaccine and translational side. Everest Medicines hired Brandon Coder, formerly of Moderna's cancer vaccine research, as VP of Translational Sciences & Discovery, and Lakshmi Srinivasan, formerly head of oncology translational medicine at Moderna, as SVP, Head of Translational Medicine, both on 2026-10-09.7
What it costs to wait for approval: Summit's balance sheet
Summit is a useful case of how much capital a late-stage oncology bet consumes before revenue. Its SEC filings show zero revenue in both FY2023 and FY2024.8 Its net loss was $614.9 million in FY2023, $221.3 million in FY2024, and $1.08 billion in FY2025, which is nearly five times the prior year's loss.8 Loss per share was $0.99, $0.31 and $1.44 across the same three years.8 Operating cash outflow grew from $76.8 million in FY2023 to $142.1 million in FY2024 and $322.9 million in FY2025.8
Cash on hand has swung widely. It was $225.3 million at the end of FY2025, $106.5 million at Q1 2026 and $419.4 million at Q2 2026.8 The filings give the levels but not the reasons for the swings, so we do not speculate on them. Capital spending on equipment is tiny by comparison: $657,000 in FY2025, up from $139,000 in FY2024 and $128,000 in FY2023.8 This is a company spending on trials and outside services, not building infrastructure, and it must cover that burn until the 2026-11-14 decision.
Commercial delivery is where money is arriving
Novocure shows the other end of the pipeline. It reported quarterly net revenues of $184 million, up 16% year over year, with 18% growth in global active patients.9 It received a CE Mark for Optune Pax in locally advanced pancreatic cancer, with Germany the first EU market to launch, and reported more than 280 active patients on Optune Pax as of 2026-06-30.9 Frank Leonard said: "This was our strongest quarter to date, with record net revenues and active patients on therapy."9
What to watch
- 2026-10-24: the ESMO presentation of detailed intismeran data, and whether Moderna and Merck announce regulatory plans.4
- 2026-11-14: the FDA decision on ivonescimab, and what it does to Summit's cash position.3
- AI claims: whether AstraZeneca or others publish measured results, such as timelines, hit rates or safety-prediction accuracy, rather than descriptions of the intended workflow. Until then, the AI thesis is a stated direction, not a demonstrated result.1
Source reliability matters throughout. The wire-service releases we cite held up on 56% of 4,960 checked claims, which is better than the MIT Technology Review piece but far from perfect, and the Summit figures come from SEC filings rather than press releases.5,9