Drug development is increasingly modular, but ultimate responsibility remains with the sponsor. In our latest article, we explore why ownership is only half the problem, and what independent institutions can build together through federated infrastructure. https://x.com/i/article/2118236681681907712
Modularity in Drug Development: Ownership Is Only Half the Problem
Drug development is increasingly organized as a network. Research, preclinical studies and clinical operations can each sit with a specialist partner, so the asset and the capabilities needed to develop it no longer have to share an organization.
Sponsors have long relied on contract research organizations (CROs) and specialist laboratories. What has changed is how much of a program can now sit outside the sponsor, and how small the sponsor can be.
As the parts multiply, the connections between them increasingly shape what the whole can do. A sponsor can contract for each part. Many of the connections depend on several organizations choosing to work in compatible ways, which one sponsor cannot arrange alone.
Two early shapes of the network
Two recent models show how differently the pieces can be arranged.
Mithril Medicine, acquired by Anthropic in July 2026, reduced the virtual biotech model to one full-time employee, supported by part-time consultants. By Anthropic's account, its founder used Claude to coordinate work across 11 external labs and took a program for a rare muscular dystrophy subtype from concept to promising mouse data in under 18 months.
Formation Bio's structure resembles the hub-and-spoke model of Roivant and BridgeBio, with a notable addition: a dedicated fund backed by outside investors alongside Formation's own capital. The $600 million fund, launched in October 2026, aims to acquire or in-license about 10 to 15 clinical-stage programs over three to five years.
Mithril reaches out to external capabilities, while Formation pools capabilities across a portfolio. Both separate where an asset is owned from where the work is organized. Each also needs something to hold that work together: a founder supported by AI in one case, a development organization serving a portfolio in the other.
One weak link can reach many sponsors
The work can spread across a network, but ultimate responsibility for the trial stays with the sponsor. Under the ICH Good Clinical Practice (GCP) guideline, in its current version, E6(R3), adopted in January 2025, a sponsor may transfer trial-related activities to service providers. It retains ultimate responsibility for participant protection and the reliability of trial data.
Two regulatory cases show how far a provider's failure can reach:
- Synchron and Panexcell. In September 2021, the FDA notified sponsors that studies conducted by these two CROs were not acceptable because of data integrity concerns. It required sponsors to repeat bioequivalence and bioavailability studies essential for approval, and an FDA spokesperson put the number of affected applications at just over 100.
- Synapse Labs. In December 2023, the EMA's human medicines committee (CHMP) recommended suspending the marketing authorizations of a number of generics tested by the CRO. A GCP inspection had found irregularities in study data, the review covered more than 400 medicines, and the recommendation was confirmed in March 2024.
In both cases, the CROs conducted the studies, while the sponsors bore the consequences: repeated studies and suspended authorizations. Both mainly involved bioequivalence studies for generics, but the principle applies to any outsourced trial, and it concerns oversight rather than geography.
The lesson reaches beyond any one sponsor. Many sponsors can depend on the same provider, so the reliability of each participant's work is a shared concern.
What the connections have to carry
Asset ownership, meaning control of an asset's intellectual property, development strategy and capital allocation, gives a sponsor the right to decide. Deciding well depends on evidence, and in a modular program much of it comes from other organizations. Every handoff can lose context or let an unreliable result through.
Once execution can be contracted, access to it becomes less of a differentiator. What increasingly matters is whether the work of many parties can be connected into a decision the sponsor can defend to regulators and investors.
Three conditions in particular determine whether evidence is fit for that decision:
- Continuity of context: each experiment, milestone and investment decision connects to the next.
- Traceable evidence: every result a decision relies on has clear provenance.
- Access to evidence held elsewhere: the sponsor can draw on evidence held by institutions it does not control.
Each condition gets harder as more organizations generate the evidence. Provenance starts with whoever produces a result, and access depends on whoever holds the record. Scientific and clinical judgment, and the authority to change course, stay with the sponsor. A stronger network can improve the information that judgment has to work with.
The network widens as a program advances
Coordination demands grow as a modular program advances, because the sponsor relies more on organizations it does not directly control.
- Discovery and preclinical work: most evidence comes from laboratories the sponsor hires and directs.
- Clinical validation: evidence comes from CROs and trial sites. Their work follows GCP standards and a study design the sponsor owns.
- Wider execution and real-world use: as work spreads across sites and countries, and once a therapy reaches real-world use, more evidence comes from hospitals and health systems. They keep their own records under their own rules.
At this point, access to the evidence, systems that can work together, and clear rules for using the evidence matter more.
Mithril suggests how AI can help with continuity of context, but its results are still preclinical. Anthropic's target for first-in-human trials is as early as late 2028, if the results hold up. Whether lean models work further along the path remains to be seen.
Cross-border deals can widen a program's network early, and Asia shows why that matters. In two 2026 analyses, McKinsey estimates that the region's share of the global innovative drug pipeline rose from 28% to 43% in five years, and that it now accounts for roughly 40% of global out-licensing transactions.
In-licensing an asset across borders means inheriting a data history generated under institutions, systems and regulations the new sponsor did not control. These programs tend to depend on outside institutions earlier and more often, because their evidence sits in more institutions, under more regulatory systems.
Connecting institutions on their own terms
A shared layer is more likely to work if institutions can join it without giving up control of their records. Connecting evidence across institutions is the problem Life AI is working on, through infrastructure designed to support the coordination of specialized capabilities and the evidence they produce. Its design–validate–deploy process is meant to keep human experts and laboratories in the loop.
The obstacle is often not only technical. Hospitals and health systems often cannot, or will not, hand source records to a sponsor or a central repository, for legal, regulatory and institutional reasons.
Life Cloud, a federated healthcare AI platform in development, is designed around that constraint. The aim is for participating hospitals to remain independent nodes and keep their source records, while the platform supports interoperability, controlled access and analysis across institutions.
The limits are part of the design. Life Cloud is meant to shorten the path when a decision is waiting on evidence held by another institution, and only where that institution chooses to take part. It cannot shorten the path when a decision is waiting on judgment, or on someone with the authority to make it. Program leadership and responsibility for clinical execution stay with the sponsor, and how far the approach can scale is a question Life AI is still working through.
Ownership is only half the problem
Modular drug development can be seen as three layers. Decisions are centralized; evidence and execution can stay distributed.
- Asset ownership sets who has the right to decide.
- A center of responsibility, whether a single founder supported by AI or a development organization serving a portfolio, applies scientific judgment and program control to turn distributed work into decisions.
- Connective infrastructure helps evidence reach that center.
For sponsors, this changes what to build and what to buy. Capabilities can be contracted, and infrastructure can make evidence reachable, but turning context, provenance and evidence into decisions that hold up with regulators, investors and partners remains the sponsor's job. For CROs, laboratories and infrastructure providers, value is likely to depend less on what they can execute and more on how well their work can be traced, connected and relied on.
Evidence can also flow the other way. What is generated in clinical care and real-world use can inform the next design decision. A development model that can draw on that evidence across institutions, while each institution keeps control of its records, should be better placed to learn from one program to the next.
In modular drug development, execution can be bought. Integration has to be built, by each sponsor within its program and by the field across institutions.
References
Anthropic, Early progress on accelerating drug development with Claude (27 October 2026): https://www.anthropic.com/research/early-progress-drug-development
The Neuron, Inside Mithril, the One-Employee Biotech Anthropic Acquired (28 October 2026): https://theneuron.ai/news/inside-mithril-one-employee-biotech-anthropic-acquired
Pulse 2.0, Formation Bio Launches $600 Million Drug Development Accelerator Fund (October 2026): https://pulse2.com/formation-bio-launches-600-million-drug-development-accelerator-fund/
Axios, Formation raises $600 million to fix AI's biotech bottleneck (29 October 2026): https://axios.com/2026/10/29/formation-bio-ai-drug-development
DCAT Value Chain Insights, Emerging Pharma: Roivant Sciences on the Rise: https://www.dcatvci.org/features/emerging-pharma-roivant-sciences-on-the-rise
Fierce Biotech, BridgeBio CEO Kumar on its hub-and-spoke model and where else it could work: https://www.fiercebiotech.com/biotech/bio-bridgebio-ceo-kumar-its-hub-and-spoke-model-and-where-else-it-could-work
ICH, Guideline for Good Clinical Practice E6(R3) (adopted 6 January 2025; consolidated version June 2026): https://database.ich.org/sites/default/files/ICH%20E6%28R3%29_Step4_FinalConsolidatedGuideline_2026_0616_.pdf
FDA, Notification to Pharmaceutical Companies: Clinical and Bioanalytical Studies Conducted by Panexcell Clinical Lab or Synchron Research Services Are Unacceptable (16 September 2021): https://www.fda.gov/drugs/drug-safety-and-availability/notification-pharmaceutical-companies-clinical-and-bioanalytical-studies-conducted-panexcell
STAT (28 September 2021): https://www.statnews.com/pharmalot/2021/09/28/fda-data-integrity-clinica-trials-synchron-panexcell/
EMA, Synapse Labs Article 31 referral (CHMP opinion December 2023, confirmed March 2024): https://www.ema.europa.eu/en/medicines/human/referrals/synapse
McKinsey, The emerging epicenter: Asia's role in biopharma's future (January 2026): https://www.mckinsey.com/industries/life-sciences/our-insights/the-emerging-epicenter-asias-role-in-biopharmas-future
McKinsey, From fast follower to originator: How Asia is rewiring biopharma innovation (24 April 2026): https://www.mckinsey.com/industries/life-sciences/our-insights/the-synthesis/from-fast-follower-to-originator-how-asia-is-rewiring-biopharma-innovation