The AI claims are the least-verified part of this story
Oncology is being reshaped by two forces at once. Personalized mRNA cancer vaccines are showing clinical validation, and large drugmakers are moving toward AI-designed biologics.1 The AI half is the harder one to pin down. Almost everything the dossier holds about it comes from one MIT Technology Review piece, "How AI helps scientists design the next generation of medicines" (undated). Our measurements show 0% of the 11 claims we checked from that source held up.2 That is a small sample, and it does not mean the claims are false. It does mean we could not confirm them. Read what follows as AstraZeneca-side intent and vision, not as demonstrated results.
What the AI-designed biologics pitch actually says
Puja Sapra, in the MIT Technology Review piece, describes a 'lab of the future' built on a self-driving-car 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."2
The stated competitive edge is data, not algorithms. Per the same source, "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."2 The article ties this to AstraZeneca's proprietary datasets covering molecular structures, binding measurements, safety profiles and manufacturing outcomes.2 For practitioners, the claim is that a large, varied internal dataset matters more than model choice. We have no verified evidence here to test that.
The same piece sets out what the models are meant to do for multi-specific biologics. They would "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."2 It also expects automation to produce "AI-ready data at a scale that traditional workflows cannot match," with robotic sample handling and integrated data pipelines speeding early development.2
The unsolved problem is safety prediction
The most useful sentence in the piece is its admission of the bottleneck: "One of the hardest problems in de novo design is predicting whether a computationally generated molecule will be safe in the human body."2 The proposed fix is virtual clinical trials that use advanced cell systems and micro-scale organ models alongside AI. These systems, it says, could "generate enhanced biological signals without traditional testing bottlenecks" and are "a critical missing piece in closing the loop between AI-generated designs and clinical-ready candidates."2 The wording is 'potential' and 'could'. The piece also says human oversight stays central, "ensuring explainable and ethical AI for the benefit of patients."2 Nothing in the dossier shows an AI-designed biologic that has cleared this step.
The mRNA track has the stronger evidence
The clearest clinical signal is a personalized vaccine from Merck and Moderna. It was reported on August 19, 2026 to have stopped melanoma from returning in a historic clinical trial.3 The companies plan to present detailed intismeran results at ESMO on October 24, 2026 and to engage with regulators on potential filings.3 Those detailed data are the real test. Until then, the headline result is reported but not yet examined in detail. This track is also a different kind of AI story: the dossier gives no evidence on how much AI contributed to these vaccines.
Commercial results: real growth, uneven sourcing
Novocure reported second-quarter 2026 net revenues of $184 million, up 16% year-over-year, driven by 18% growth in global active patients.4 Frank Leonard called it "our strongest quarter to date, with record net revenues and active patients on therapy."4 The company also received a CE Mark for Optune Pax in locally advanced pancreatic cancer, with more than 280 active patients as of June 30, 2026.4 This source's measured reliability is 57% of 4,954 checked claims holding up.4
The wider narrative says Q2 guidance was raised at Hematology and Oncology, Opzelura and Kiniksa.1 The dossier contains no underlying document for those three, so we cannot confirm the figures and have not tried to supply them.
Other clinical and regulatory markers that appear in the dossier:
- The CHMP adopted a positive opinion on Gilead's Trodelvy plus Keytruda in first-line PD-(L)1-positive metastatic triple-negative breast cancer, on July 24, 2026.5
- Telix dosed the first patient in the Phase 3 LUTEON trial of TLX250-Tx in renal cancer, described as the first radiopharmaceutical therapy to enter Phase 3 for clear cell renal cell carcinoma.6
- The FDA PDUFA goal date for Summit's ivonescimab application, which is based on HARMONi trial results, is November 14, 2026.3
Small biotechs: cash is the constraint
Funding pressure remains at smaller companies. ADC Therapeutics says it "maintains a cash runway into 2028, supporting ongoing regulatory and clinical plans for ZYNLONTA combinations."7 The same weekly roundup headline also notes job cuts at ADCT.7 Treeline is taking a different route. Its transaction with Standard BioTools is expected to leave the combined company with over $900 million at closing.8 The release reports Phase 1 TLN-121 monotherapy data in relapsed or refractory lymphomas, with an overall response rate of 84%, a complete response rate of 32% and no observed dose-limiting toxicities.8 Planned 2027 data updates are listed for TLN-121 and TLN-372.8 Josh Bilenker said the cash "will help us accelerate the development of our clinical programs."8 That source's measured reliability is lower: 31% of 2,925 checked claims held up.8 Treat the response rates as company-reported.
What big pharma spending shows, and what it does not
Merck is the one company with SEC-verified numbers in the dossier, and none of them isolate AI or oncology spending. Its capital expenditure was $3.372 billion in fiscal 2024 and $4.112 billion in fiscal 2025, an increase of roughly 22%.9 Year-end cash was $6.841 billion in 2023, $13.242 billion in 2024 and $14.565 billion in 2025.9 By the second quarter of 2026 cash stood at $6.849 billion, against $8.007 billion a year earlier.9 Quarterly cost of revenue rose to $4.395 billion in Q2 2026 from $3.557 billion in Q2 2025.9 We compare these only with Merck's own history, because the dossier offers no outside yardstick. They show a company spending more and carrying large cash swings. They do not show what portion goes to AI or to intismeran.
What to watch
- October 24, 2026: detailed intismeran data at ESMO, and any word on filings.3
- November 14, 2026: the FDA decision date for ivonescimab.3
- AI evidence: whether AstraZeneca or anyone else publishes a measured result, such as a safety-prediction accuracy figure or a candidate that moved from AI design into trials. Until then the AI case rests on stated ambition.
- 2027: planned TLN-121 and TLN-372 data updates from Treeline.8