In THIS ISSUE
Signal of the Week, AI deal headlines promise billions, commitments are millions.
Framework of the Week, The Option Test: four checks before you believe an AI deal headline.
5 Things Worth Reading, deals, regulators, and the MSL shift.
Tool of the Week, Open Targets Platform.
Reading time: ~8 min read
🎯 Signal of the Week
Two AI drug deals this week, and in both the cash committed today is under 3% of the headline.
Welcome to Issue 12,
Forty million dollars. That is the cash AstraZeneca commits to insitro today, in a collaboration announced on October 15 with a headline of up to $1.5 billion. The remaining $1.46 billion or so depends on development and commercial milestones, plus single-digit royalties on global sales. insitro applies its Virtual Human platform to find and advance targets in MASH and cardiomyopathies up to a development candidate. AstraZeneca nominates the targets, runs clinical development, and holds exclusive worldwide rights.
It was not the only one. Two days earlier, on October 13, Zealand Pharma agreed to pay Generate Biomedicines $30 million upfront, with early preclinical milestones of up to $35 million and nearly $1.2 billion in total milestones, for AI-designed multispecific proteins in obesity and other metabolic diseases. In both deals the upfront is under 3% of the headline. For scale, AstraZeneca spent $13.85 billion on R&D in 2025, so $40 million is about 0.3% of that. Yahoo Finance's analysis put it bluntly: an option, not a bet. Read the caveats too: neither announcement says how much could become payable before a candidate reaches the clinic, and the words "only $40 million is guaranteed" are the publisher's framing, not the companies'.
Figure 1. The upfront is the commitment: cash committed now versus headline value in two AI drug deals announced in October 2026. Sources: Yahoo Finance; StreetInsider; Contract Pharma.
Here is why this belongs on a Medical Affairs agenda. Discovery deals feel far from your desk, and they are not. The moment a target becomes a candidate, AstraZeneca takes over clinical development, regulatory work, and commercialization, and the evidence you will have to explain later is being chosen now. Both MASH and cardiomyopathy are heterogeneous: one diagnostic label covers patients with different underlying biology, which is the problem insitro says its platform is built for. MASH also has a long record of expensive target mistakes, with obeticholic acid, elafibranor, selonsertib, and emricasan all failing before resmetirom and semaglutide were approved. How you later explain who benefits starts with how the target was picked.
Now the opinionated part. Treat a headline value as a ceiling on optimism, not a measure of belief. A 2.5% upfront means each buyer paid for the right to wait for evidence, which is sensible. It is also how you should read AI claims inside your own function: the vendor's headline is the milestone total, and your real commitment is the upfront. Last week's issue asked who let the agent in. This week's asks what has actually been committed once it is, and what has to be true before the next payment.
🧠 Framework of the Week
The Option Test: four checks before you believe an AI deal headline.
How much of a billion-dollar headline is a promise? Usually very little, and that is fine, as long as you can say which part is which. The two deals above are well-built options: small cash now, large payments only if the science clears defined gates. The risk sits with the reader who treats the option price as a forecast.
This builds on the thread from recent issues. We followed the agent into the workflow, the system of record, the shared rental, the exam room, and past the badge office. This week we apply the same discipline to money, because deal news is where AI claims are loudest and where Medical Affairs is most often absent. When the evidence plan is written after the deal is signed, you inherit questions you could have shaped.
Read the framework as four questions in order. Checks 1 and 2 are for anyone who reads deal news, and they take two minutes. Checks 3 and 4 are where Medical Affairs adds value, because you know which evidence the field will demand later. The same test works on a vendor contract for your own AI tools: swap the buyer's R&D budget for your function's budget.
Figure 2. The Option Test: four checks, from the first payment to the evidence you will own. Source: iCHealth Pathway framework (original).
Cash now. Separate the upfront from the headline, then express it as a share of both the headline and the buyer's R&D budget. AstraZeneca and insitro: $40 million of $1.5 billion, which is 2.7%, and about 0.3% of 2025 R&D spend. My rule of thumb: under roughly 5% of the headline, treat the deal as an option on evidence, not a commitment to an outcome.
First trigger. Find the first milestone that releases real money, and what has to be true for it to pay out. In discovery deals that is usually a development candidate or a first preclinical result; the Zealand and Generate deal names early preclinical milestones of up to $35 million. If you cannot find the trigger in the announcement, you cannot yet judge how serious the buyer is.
Handover. Ask who owns the program once the AI partner's work ends. Here insitro stops at the development candidate and manufacturing planning, and AstraZeneca leads clinical development, regulatory work, and commercialization. Everything after the handover is your company's problem, so list the questions Medical Affairs will inherit: patient subgroups, endpoints, comparators, and natural history.
Evidence you own. Write down the evidence that will exist at handover and the evidence that will not, then ask for an evidence plan at nomination instead of at Phase 3. In a heterogeneous disease, that means deciding early how patient subtypes will be defined and described. Engage when targets are chosen, because that is when the cheapest changes are still available.
The first two checks tell you how serious the buyer is. The last two tell you what you will be asked to explain. Most readers stop at the headline and the upfront. Keep going to the handover, because that is where an AI discovery deal turns into a Medical Affairs workload, and treat the milestone total as a hope, not a forecast.
Figure 3. The size of the bet: AstraZeneca's deal with insitro set against its own 2025 R&D spend, with the maximum and upfront values drawn to scale. Sources: Yahoo Finance; PharmTech.
A headline is what a deal could be worth. The upfront is what the buyer believes today.
📚 5 things worth to read
Five paywall-free reads, each summarized so you get the value without the click
01
MASH and Cardiomyopathy Targets: Why AstraZeneca Chose insitro's AIThe closest look at the science and the operating consequences of this week's signal. Both diseases are heterogeneous, one label covering patients with different underlying biology, which is the case for a platform that pairs large human and cellular datasets with AI. It reminds readers that MASH affects more than 250 million people and that several candidates failed before resmetirom and semaglutide were approved. It quotes AstraZeneca's Sharon Barr on target selection being one of R&D's most critical decisions. The practical caveat is the best part: insitro's role ends at the development candidate, so faster candidate flow may strain development and manufacturing teams downstream. The line to keep: speed at the front of the pipeline moves the bottleneck to everyone behind it, Medical Affairs included.
02
FDA Seeks Public Feedback on Regulatory Approach for Generative AI-Enabled Medical Devices
A law firm's clear summary of the FDA's August 18 discussion paper, which is nonbinding and not guidance. It proposes a two-axis risk view, weighing how independently a function acts against how severe the harm from a wrong output could be, and a competency-based premarket path that pairs benchmarking with clinical confirmation. It asks whether more premarket uncertainty could be accepted if monitoring, re-benchmarking, and drift detection are strong, and floats a voluntary Foundation Model Device Master File and agentic AI oversight. Comments close on October 19 under docket FDA-2026-N-7874, so today is the last day. The line to keep: regulators are moving from approving a model once to supervising it for life, which is the Option Test's logic applied to evidence.
03
Digital Omnibus Adopted: AI Act High-Risk Delayed to 2027A tracker for the EU's Digital Omnibus on AI, now Regulation (EU) 2026/1744, published in the Official Journal on July 24 and in force since July 27. High-risk obligations for standalone systems move from August 2026 to December 2, 2027, and for AI embedded in regulated products such as medical devices from August 2027 to August 2, 2028. Article 50 transparency has applied since August 2, 2026, with a grace period until December 2, 2026 for machine-readable marking, and only for systems already on the market. Penalty ceilings are unchanged, up to 35 million euros or 7% of turnover. It is a commentary site, so confirm details against the Official Journal text. The line to keep: the dates moved, the obligations did not.
04
Webinar Recap: The Future of Field Medical, How AI Is Reshaping MSL Workflows and HCP Engagement
A June recap of a MAPS and Sorcero webinar with speakers from Gilead, MSL Mastery, and Sorcero. The headline numbers: after AI training, MSLs reportedly saved about 90 minutes of pre-meeting preparation per KOL, around 45 hours across a 30-KOL panel, and roughly 20 minutes of writing per insight. Gilead's Byron Burnette said the goal is higher quality work, not more work. The panel also described using AI to aggregate signals across a region, and building small agents such as an email digest or a devil's advocate that tests scientific arguments. It is a vendor-published recap and the savings are self-reported, not independently measured. The line to keep: spend the saved hours on sharper questions, not on more meetings.
05
Is Medical Affairs About to Become Obsolete as a Result of AI?
An opinion piece by Samuel Dyer, CEO of the MSL Society, published in July 2025 and still the most cited framing of the job-market worry. Using LinkedIn polls, it reports that 72% of 806 respondents named Medical Information as the role most likely to shrink significantly, 55% of 520 expect leadership reductions, and 69% of 136 say finding a new Medical Affairs role is harder than before. Dyer concludes the function stays essential with fewer positions, and that empathy, judgment, and trusted KOL relationships are the durable skills. The polls are small and reflect perception, and the piece is more than a year old, so read it as a mood, not a measurement. The line to keep: the work is changing form, so pick the part of it a machine cannot sign for.
🔧 TOOL OF THE WEEK
Open Targets Platform
What it is: a free, open platform that integrates publicly available datasets to help scientists identify and prioritise drug targets. It comes from the Open Targets partnership around EMBL-EBI and the Wellcome Sanger Institute. For any disease or gene you get ranked associations built on genetic evidence, expression, known drugs, pathways, and the literature. The June 2026 release added single-cell and proteomics expression data, more than 14,000 new GWAS studies, and 292,325 clinical precedence evidence items from mined trial records. Its data are released under CC0, and the code, including a GraphQL API, is open source under Apache 2.0.
How to use it for Medical Affairs and MSLs: when a target deal lands, such as AstraZeneca and insitro in MASH, search the disease, look at which targets rank highest and what kind of evidence backs them, and note where the genetic support is strong or thin before your next internal strategy meeting; use the clinical precedence evidence to see which targets have already been tested, which helps you anticipate competitor positioning and questions from HCPs; and, as a companion to this week's Evidence you own check, pin two or three targets side by side and share the page link, because the platform keeps filters and pinned targets in the URL, so a colleague opens exactly what you saw.
Where to access it: platform.opentargets.org on the web, free to use. Documentation, licence terms, the GraphQL API, and bulk data downloads are described at platform-docs.opentargets.org. Release notes appear on the Open Targets blog, and the platform changes with each release, so confirm current features before you build a process around it.
COMPLIANCE NOTE
Open Targets associations show evidence that a target is linked to a disease, not that a drug against it will work, so never present a ranking as efficacy or as MLR-approved content. Scores depend on data sources and release versions, and a literature pipeline update in June 2026 removed about 60,000 direct associations with weaker support, so record the release number whenever you cite it. Individual data sources carry their own terms, even though the platform dedicates its data to the public domain. Do not paste confidential pipeline plans, unpublished data, or patient-identifiable information into any search or shared link, and check your company's policy on third-party tools before using it on a work device.
That is Issue 12. If a colleague forwarded this to you, you can get it yourself every Monday at newsletter.ichealth-ai.com/subscribe.
One email a week. No sponsorship. See you next Monday.
iCHealth Pathway is a weekly research note by Dr. Issam Chebouti on AI transformation for Pharma, Medical Affairs, and Healthcare leaders.
iCHealth-AI, ichealth-ai.com, [email protected]
Issue 12, Monday, October 19th, 2026.
Parts of this issue were produced with AI assistance and checked before publication.