The AI claims are forecasts, and the source says so
The most AI-relevant material this month is an interview-style piece in MIT Technology Review featuring AstraZeneca's Puja Sapra, on designing medicines with AI.1 It is an articulate statement of where a large pharma company thinks the field is heading. It is not a report of a drug designed by AI and proven in patients. Our measured reliability check on that document found 0% of 11 checked claims held up.1 That does not mean Sapra's statements are false. It means we could not confirm them against an independent source, so we treat them as stated intent.
The central bet is that data, not model architecture, is the moat. Sapra says: "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 The practical meaning for AI practitioners is that the competitive claim rests on proprietary, multimodal data the company already owns, not on a model nobody else can build.
What the 'agentic' pitch actually describes
The piece describes a shift toward systems that generate candidate molecules and predict their efficacy and safety together. Sapra puts it this way: "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 wording is conditional ("could help"), and it should be read that way.
The 'lab of the future' is the most concrete part. Sapra uses an analogy: "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 She argues this "will generate AI-ready data at a scale that traditional workflows cannot match," and that robotic sample handling, automated quality checks and integrated data pipelines "have the potential to help accelerate early drug development timelines significantly."1 The loop being described is design, then automated experiment, then data, then redesign. The word 'potential' is again doing real work.
The hard part is named in the same piece
The most useful sentence for a technical reader is a limitation. Sapra says: "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 built on 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 By her own account, then, the loop between AI-generated designs and clinical-ready candidates is not yet closed.
She also keeps people in the picture: "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 And: "With more autonomous systems, human oversight remains at the heart of this approach—ensuring explainable and ethical AI for the benefit of patients."1 The framing is AI that speeds up scientists, not AI that replaces the validation step.
Meanwhile, the near-term oncology news involves no AI
The regulatory events in this cycle are conventional drug development. On October 9, 2026 the FDA approved Tecentriq and Tecentriq Hybreza, combined with a fluoropyrimidine and oxaliplatin, for adjuvant stage III dMMR colon cancer, based on the Phase III ATOMIC study.2 On October 16 the EU's CHMP adopted a positive opinion on an alternative Libtayo schedule: 700 mg as a 30-minute IV infusion every six weeks after an initial 12 weeks of 350 mg every three weeks.3 Summit's ivonescimab application, based on HARMONi trial results, has an FDA goal date of November 14, 2026.4 Moderna and Merck plan to present detailed intismeran results at ESMO on October 24.5 We found no AI-designed molecule among these. The dossier does not say how any of these products were discovered, so we draw no conclusion about their origins.
Talent movement is one small signal that AI-adjacent work is attracting hires. On October 9, Everest Medicines appointed two former Moderna oncology translational scientists: Brandon Coder, formerly Research & Early Translational Lead for Cancer Vaccines, and Lakshmi Srinivasan, formerly Head of Oncology Translational Medicine.6 The dossier does not say whether AI drove these hires, and we do not claim it did.
The cost of the conventional route: Summit's filings
For a sense of what a late-stage oncology programme costs before any AI shortens it, Summit Therapeutics' SEC-verified figures are instructive. It reported zero revenue in both FY2023 and FY2024.7 Its net loss was $614.9 million in FY2023, $221.3 million in FY2024 and about $1.08 billion in FY2025 (-$1,079,586,000), roughly five times the prior year's loss.8 Cash used in operations rose from $76.8 million in FY2023 to $142.1 million in FY2024 and $322.9 million in FY2025.9 Earnings per share moved from -$0.99 to -$0.31 to -$1.44 over the same three years.10
Its cash balance has swung widely. It stood at $225.3 million at the end of FY2025 and $106.5 million in Q1 2026, then $419.4 million in Q2 2026.11 The filings give the balances but not the reason for the Q2 2026 rise, so we do not speculate on it. Our own narrative analysis ties this to a wider pattern of capital consolidating around oncology services and combinations, including McKesson and CD&R's acquisition of Option Care Health and the Summit, AstraZeneca and Daiichi Sankyo collaboration on ivonescimab plus Datroway in first-line TNBC.12 Those deal terms are not in our verified dataset, so treat that as our reading, not verified fact.
What to watch
- November 14, 2026: the FDA goal date for ivonescimab. This is a regulatory test of a conventionally developed asset, and Summit's filings show how much is riding on it.4
- October 24, 2026: the ESMO presentation of intismeran data, and any discussion of regulatory filings.5
- Evidence, not vision, on AI design: a named, AI-designed biologic with disclosed clinical data would move this from forecast to result. Nothing in the current dossier does that, and the safety-prediction problem Sapra herself calls one of the hardest is the likely gate.1