Designing responsible AI and robotics for future medicines operations

Automation is not autonomy

Robotics can move materials, repeat physical tasks, support instrument loading and improve consistency. Software can schedule work, monitor signals, detect patterns and support review. Artificial intelligence may help in defined analytical or operational contexts.

None of those tools becomes trustworthy merely because it is automated. The process still needs an intended use, accountable owner, defined inputs and outputs, performance criteria, operating limits, exception handling, records, change control and human authority.

A glossy visual of a robot beside laboratory glassware provides no evidence of those controls. On the VARUNÉ Labs site, such imagery must remain explicitly conceptual.

Begin with context and risk

The NIST Artificial Intelligence Risk Management Framework organises AI risk work around four functions: Govern, Map, Measure and Manage. It is voluntary and not specific to medicines. It cannot replace United Kingdom regulation, pharmaceutical quality requirements or validation for a particular use.

Its value at concept stage is structural. Governance establishes responsibility. Mapping defines context, users, affected parties and possible harm. Measurement tests performance and uncertainty. Management sets priorities, controls and response.

This sequence helps resist a common error: choosing an impressive technology first and searching for a use later.

Good AI practice is life-cycle practice

In January 2026, the FDA and EMA published guiding principles of good AI practice in drug development. The principles emphasise human-centred design, risk-based oversight, a clear context of use, suitable expertise, traceable data governance, performance assessment, life-cycle monitoring and clear communication of limitations.

These principles do not approve any VARUNÉ system. They offer a useful lens for a future operating model.

An AI system may change because its data, environment, user behaviour, dependencies or model change. Control at launch is not enough. The organisation needs to know which version acted, what data informed it, how output was used, how performance is monitored and when a person must intervene.

Robotics creates a physical evidence problem

Robotics adds physical movement to digital risk. A proposed environment must consider access control, guarding, collision, contamination, cleaning, material identity, chain of custody, instrument state, calibration, recovery and safe manual operation.

The best first use may be repetitive and bounded. A robot that reliably transfers identified items under controlled conditions may create more value than a complex system attempting to direct an entire experimental programme.

Modularity matters. A failure in one cell should not make the whole environment unintelligible or unsafe. Interfaces, states and handovers should be visible to the people accountable for the work.

Sector interest is not proof of readiness

The Innovate UK competition Medicines Manufacturing Labs of the Future sought projects involving digital, automated and robotic technologies for pharmaceutical process development and manufacturing. The competition opened in April 2026 and closed in May 2026.

This is evidence that the area is of public innovation interest. It is not evidence that VARUNÉ applied, received an award, joined a consortium or owns any resulting technology.

Similarly, the MHRA Innovation Office provides a route for questions about innovative medicines and novel manufacturing approaches where regulatory uncertainty exists. Referring to that route does not imply engagement, advice or approval.

A proposed control model for VARUNÉ Labs

Any future VARUNÉ intelligent system should pass a documented sequence before operational use.

1. Define the decision

   State the task, user, context, consequence and expected benefit. Identify decisions the system must never make.

2. Establish data authority

   Identify source, permission, quality, lineage, retention and exclusions. Sensitive or regulated data should not enter a system merely because access is technically possible.

3. Design human control

   Name the accountable owner, required review, override, escalation and safe state. Human oversight must be operational, not ceremonial.

4. Test performance and failure

   Set relevant measures, acceptance criteria and challenge cases. Test drift, missing data, unusual inputs, physical faults and dependency failure.

5. Control deployment and change

   Record versions, approvals, configuration and release state. Monitor performance in use and define when the system must pause, roll back or be retired.

6. Preserve an audit trail

   Keep enough evidence to reconstruct what happened, which system acted, what a person reviewed and why a decision was accepted.

The proposed intelligent systems environment

VARUNÉ Labs currently presents an intelligent systems environment as a future campus concept. It may explore laboratory robotics, connected equipment, operational simulation and decision support. No system is represented as built, validated, deployed, autonomous or accepted by a regulator.

The purpose of the concept is to connect technology ambition with governance from the beginning. The strongest future system will not be the one that appears most autonomous. It will be the one that performs a valuable task, exposes its limits and remains under clear human authority.

Conclusion

Responsible automation is an operating discipline. It combines a useful context, traceable evidence, physical and digital controls, accountable people and continuous review.

For VARUNÉ Labs, artificial intelligence and robotics remain proposed areas of future capability. Their credibility will depend on what can be demonstrated safely and repeatably, not what can be rendered convincingly.