MSD and Google Cloud: What AI agents mean for pharma
4 min read
US pharmaceutical company Merck & Co., known as MSD outside the US and Canada, plans to deploy a platform for AI agents across several business areas with Google Cloud. The multi-year partnership covers research and development, manufacturing, sales and corporate functions. For mid-sized pharmaceutical and chemical companies, the announcement offers pointers for planning their own projects. The specific use case, access to data and responsibility for results are key.
Key points at a glance
- MSD and Google Cloud plan deployment across business areas. Google Cloud teams are to work with employees of the pharmaceutical company.
- The first use case needs clear boundaries. Companies should define which data an agent may access and who checks its results.
- Legal obligations depend on the specific use. The obligation to promote AI literacy already applies. A system is not classified as high-risk solely because it is used in the pharmaceutical industry.
What MSD and Google Cloud announced
The partners announced their collaboration on 22 April 2026. According to the company announcement, the platform is to use Gemini Enterprise among other technologies. Google Cloud engineers are working with MSD teams on the project. Applications are planned in research and development, manufacturing, sales and corporate functions.
The announcement describes objectives and intended areas of use. It does not yet reveal what results the individual applications will achieve in operation. The 75,000 employees mentioned there refer to the pharmaceutical company’s global workforce and do not establish that a rollout to every person has been completed.
The name matters when considering the announcement in Germany: US company Merck & Co. operates as MSD outside the US and Canada. Germany’s Merck KGaA in Darmstadt is a separate company.
From business process to first agent
What is an AI agent? An AI agent is a system that can carry out several steps to achieve a goal. Depending on how it is configured, it might search for information, prepare a proposal or use a connected tool. The individual application determines which steps it may carry out independently.
For a mid-sized company, a clearly defined process is a suitable starting point. In quality assurance, for example, an agent could compile documents relating to a deviation and prepare a draft for the specialist responsible. Whether the sources are complete and which measures should follow remain subject to specialist review.
Before a pilot project, the business department and IT should jointly define what data is needed, who may access it and how results will be checked. This also includes a procedure for erroneous outputs: who corrects them and when is the agent stopped? Such rules provide the basis for assessing whether deployment makes sense in the intended process.
Review data and providers in contracts
Research documents, production data and sales information have different protection requirements. For each planned use of data, companies should therefore check its origin, who is authorised to access it and its intended purpose. Where personal data is involved, the preparation should include a data protection assessment.
When evaluating offers for an agent platform, data processing, access rights and responsibilities matter alongside price. Procurement and IT can establish how usage is billed, which logs are available and how data and configurations can be exported when changing providers. The offer should state whether a service provider uses funding and whether the customer benefits from it.
Google Cloud has announced a fund for its global partner ecosystem. This does not give an individual customer an entitlement to a grant or a particular project price.
Clarify legal requirements and responsibility
The EU AI Act applies in stages. According to the European Commission, the obligation for providers and deployers to take measures to promote the AI literacy of people involved in operating and using AI systems has applied since 2 February 2025. The measures should take account of the knowledge of the people involved and the context of use, among other factors.
Whether an AI system is classified as high-risk and which obligations follow depends on the specific system, its purpose and the company’s role. Use in pharmaceutical production alone does not answer that question. For applications close to production or relevant to quality, companies should also assess the applicable sector-specific requirements separately.
Logs can help make actions traceable. Companies should check before deployment what information needs to be recorded and whether a product actually provides it. Calling an application an AI agent does not in itself create an audit trail.
Frequently asked questions
Which Merck is involved in the partnership?
The company involved is US pharmaceutical company Merck & Co., known as MSD outside the US and Canada. Germany’s Merck KGaA is a separate company.
Must AI agents in pharmaceutical production be classified as high-risk?
The industry alone does not determine this. The specific system, its intended purpose and the applicable rules are decisive. Companies should assess each individual use.
What should mid-sized companies define before a pilot project?
This includes the defined process, permitted data and actions, specialist review of results and a procedure for correcting or stopping the agent.
Read more on MyBusinessFuture
MyBusinessFuturePharmaceutical serialisation next level: EU FMD revision for IT and supply chainsMyBusinessFutureAI agents in ERP: Autonomous processes with SAP, Microsoft and OracleMyBusinessFutureAI Reskilling 2026: Key Roles in German Mid-Sized FirmsMore from the MBF Media Network
cloudmagazinAWS and Google Cloud multicloud interconnect reaches general availabilityDigital ChiefsAutodesk appoints a16z CIO Mike KellySecurityTodaySecurity data fabric for mid-sized businessesImage source: AI-generated (November 2026)
Translated from the German original using artificial intelligence. The German version is authoritative.
