Repurgenesis signs AI drug MOU, but milestones remain undisclosed

Taiwan-based Repurgenesis said in late July 2026 that it had signed a memorandum of understanding (MOU) with Switzerland-based Medvisis to explore AI-assisted drug repurposing. The company's public statement says the planned collaboration would combine its Intelligent Orchestrator platform with Medvisis's European clinical and pharmaceutical network to pursue validation and commercialization opportunities.

The statement establishes an intended partnership, but it does not identify a drug candidate, dataset, model result, laboratory program, study sponsor, budget, deadline or licensing term. APPI News could not find the memorandum itself or a binding development agreement in the public material reviewed at the time of writing.

The first test is a dated work plan

Because the MOU text is not public, APPI News could not assess whether any provision creates binding obligations. The announcement does not establish that either company has committed to fund work, transfer data, deliver a candidate or negotiate a later agreement.

A usable work plan would name the candidate or disease area, the intended AI task, the party responsible for each stage, the deliverable, the deadline and the conditions for stopping. It would also separate a completed event, such as signing an agreement, from a company target, such as finding a new use or securing a license.

Failure criteria matter because a work plan must cover an unsuccessful result as well as an advance. A contract can specify what happens when a model misses a pre-agreed threshold, laboratory work does not reproduce the prediction or a safety signal changes the program's risk assessment.

Data quality comes before a model score

The intended context of use should be defined before performance is assessed. The European Medicines Agency and the US Food and Drug Administration published 10 joint principles in 2026 covering context of use, data governance, model development, risk-based performance assessment and life-cycle management.

Those principles are not a certification of any company or platform. They provide a framework for questions about a system, while evidence for a particular model still has to match the task, data and decision in which it will be used.

The European Medicines Agency's 2024 reflection paper says sponsors, applicants or manufacturers are responsible for ensuring that algorithms, models, datasets and processing pipelines are fit for purpose. It also calls for data sources, acquisition and processing steps to be documented in a detailed, traceable way and for representativeness and bias to be examined against the intended task.

For this partnership, a useful data disclosure would identify the source, time period, population, permitted uses, missing-data treatment, labels and version of each dataset. It would also show that training, validation and test records are separated in a way that prevents the same person, sample or closely related record from leaking across those sets.

The announcement refers to a European network but does not say that personal or health data will cross a border. APPI News therefore could not determine which data-protection regimes would apply. Any later project would need to identify the data controller, storage location, access rights, transfer basis, retention period and deletion process.

Independent testing must match the intended claim

An internal model score can show performance on a defined test, but it cannot by itself establish that a candidate is safe, effective or ready for clinical use. A reportable validation package would state the comparator, endpoint, acceptance threshold, confidence interval, test-set characteristics and model version, then show whether an independent team can reproduce the result.

At the discovery stage, a ranked candidate or proposed new indication remains a hypothesis. Laboratory work, nonclinical evidence and human research may follow, depending on the candidate, the intended claim and the requirements in each jurisdiction.

The US Food and Drug Administration's January 2025 draft framework applies when AI produces information used to support US regulatory decisions on a drug's safety, effectiveness or quality, and the notice expressly excludes drug discovery. It does not validate a discovery platform or a candidate, although its context-of-use and risk-based credibility approach could become relevant if AI output later enters a US regulatory submission.

Roles become testable when work enters a clinical trial

The public statement does not assign responsibility for data control, model development, laboratory testing, clinical operations, regulatory contact or licensing. A formal agreement can allocate those tasks and set ownership of background intellectual property, jointly created results, model improvements and negative findings.

The International Council for Harmonisation's 2026 E6(R3) guideline says clinical-trial roles and responsibilities should be clear and documented, investigators retain ultimate responsibility for delegated activities, and sponsors retain ultimate responsibility for transferred sponsor activities. That rule concerns clinical trials; it does not show that this MOU has produced a trial or named a sponsor.

If a human study begins, the external record should identify the sponsor, protocol, ethics review, registry entry, endpoints and participating institutions. A conference presentation or company announcement cannot substitute for those records.

Regulatory documentation is different from publishing source code

Switzerland's drug regulator, Swissmedic, says opaque AI models create challenges in data-quality assessment, model transparency, quality control and bias, and that authorization applications must contain complete documentation consistent with current science and technology. The agency's statement describes what an applicant may need to provide for regulatory assessment in Switzerland, not proof that the Repurgenesis platform is under review.

Regulatory access and public disclosure serve different purposes. A company may protect source code and trade secrets while still keeping traceable records of the model version, intended use, data scope, performance, limitations, changes and human review available to partners and regulators where required.

Six records would show whether the partnership has advanced

1. A formal agreement should identify the work scope, responsible parties, deliverables, deadlines, funding and termination conditions.
2. A candidate record should name the molecule or asset, the proposed use and its existing regulatory status in each relevant country.
3. A data and model file should document provenance, permitted uses, dataset splits, versions, comparators and acceptance thresholds.
4. A validation record should report independent-data results and any laboratory or nonclinical evidence, including negative findings.
5. A clinical record should identify the sponsor, ethics review, trial registration and regulatory interaction if work reaches human research.
6. A licensing disclosure should name the asset, territory, rights granted and payment conditions rather than describe negotiations as a completed transaction.

A dated, attributable record at any of these stages would allow the partnership's progress to be checked. Until one appears, the public evidence supports only that the companies intend to collaborate around AI-assisted drug repurposing.