Despite surging GenAI adoption in pharma and the FDA’s recent embrace of the technology, a trust gap remains with GenAI implementations. The story describes four essentials that improve trust in GenAI so that pharma and biotech teams can better navigate the present time of uncertainty.
Pharma and biotech organizations are facing a level of uncertainty in 2025 that far exceeds that of a year ago. Tariffs, global supply chain concerns, threat of recession, patent expirations, impact of the Inflation Reduction Act (IRA), changing regulatory requirements like European Commission’s Joint Clinical Assessment (JCA) framework all create an unstable environment for drugmakers.
During times of upheaval, when finely honed revenue streams appear threatened, executives rely on strategies to optimize their high-value talent and do more with less. Generative artificial intelligence (GenAI) is often hailed as the answer, because GenAI systems can be trained to “learn” how to generate content and “think” with human-like fluidity while performing mundane yet important tasks.
GenAI traffic surged 890% during 2024 across a sampling of over 7,000 organizations in various sectors, with organizations having an average of 66 GenAI apps in their infrastructure. Over 85% of these use cases are in four use cases:writing assistants, conversational agents, enterprise search, and developer platforms.1
Pharma is no different, with 49% of pharma and biotech companies using some form of AI in 20242 and three out of four industry respondents stating they were either using, testing, or actively exploring AI in their operations to meet related goals.3 Early GenAI uses in pharma have demonstrated promising results where AI helped teams streamline time-consuming functions—all important for achieving greater flexibility in market response.
At the same time, a trust gap exists where leaders may learn of hallucinations elsewhere. They may realize the complexity of harmonizing disparate data sources and want more assurances before moving from proof of concept to deployment. Given that new treatments target human health and wellbeing, caution is understandable.
The key is in realizing where GenAI fits, where it doesn’t, and structuring it to build trust and from it, confidence, and success.
The US Food and Drug Administration (FDA) underscored GenAI’s importance when, in June 2025, FDA commissioner Marty Makary announced the launch of Elsa4, a GenAI tool that is being used to expedite clinical protocol reviews and reduce the overall time it takes to complete them. During early use, Elsa helped FDA team members to summarize adverse events, conduct expedited label comparisons, generate code for non-clinical database development, and help inspectors identify high yield inspection targets. According to Makary, FDA internal staffers completed tasks with Elsa that formerly required two to three days, in just six minutes.
While the FDA's embrace of GenAI and Elsa remove an element of uncertainty as to Federal AI policy, given the January 2025 suspension of the previous administration AI frameworks, there is still plenty unclear as to what the FDA will accept in AI-generated output in regulatory submissions. Newer, less established technologies, like GenAI, may raise regulators’ questions where there otherwise were none.
Add to that the dizzying evolution of GenAI which is causing IT leaders to throw out the playbooks of a year ago with AI models such as Gemini, ChatGPT, or ClaudeGPT changing monthly.
As AI use matures beyond proof-of-concepts, pharma companies are finding, as did the FDA, that GenAI implementations include far more than bots, and indeed can be “force multipliers” with such significant time- and costs saving that they provide a decided competitive advantage. Three applications in pharma have emerged as areas where teams have moved from cautious optimism to quantified results with GenAI.
The key is knowing how to build AI infrastructure that addresses transparency, accuracy, security and privacy, and human guidance in a way that builds trust. This foundation comes from teams understanding the critical human role in both the development and continued oversight of AI systems in ensuring their accuracy and fidelity—ultimately providing confidence and added value for users.
Many organizations have developed internalAI disclosure and responsible use policies to help avoid potentially serious AI inaccuracies or mishaps.Relevant stakeholders should always have access to appropriate information around AI systems, including disclosure of when AI is used for specific tasks or decisions, traceability of AI use, justification of the accuracy and fairness of automated recommendations, and explainability on AI decision-making. More specific data governance standards6 for each use case provide further assurance that GenAI conclusions hold validity.
Teams should establish an evaluation methodology and metrics for assessing accuracy and overall performance of AI systems. If a gold-standard ground-truth dataset is not available, it should be curated. This dataset, along with the evaluation metrics, will form the benchmark to measure accuracy improvement for every model iteration. Taking a systematic approach to assessing each metric allows teams to evaluate AI-enabled output and understand the areas that require more human intervention and validation.
Deploying the most current security frameworks, guardrails, and governance processes is essential. Any pharma AI infrastructure should follow zero trust7 principles, so that each AI request or action is verified before execution. They also must demonstrate compliance to data privacy regulations like GDPR, HIPAA, CCPA, and the EU AI Act which governs the handling of patient and personal data.
Most pharma applications validate GenAI conclusions by using a Human-in-the-Loop (HITL) approach where subject matter experts review GenAI-made decisions to make sense. In this way teams work to mitigate potential bias and improve accuracy. HITL practices also help in responding to security threats, so that something beyond the automated system can decide whether security measures need to be adjusted.
Trust in GenAI is paramount to enabling teams to rely on AI-enabled systems. Once pharma leaders help to assure GenAI accuracy, security, and privacy, they set the stage for enabling their applications to disrupt existing lengthy processes. They can de-risk a host of changing variables and enable teams to fine-tune existing AI solutions into engines of organizational flexibility. That’s a strategy that transcends any specific implementation during this time of uncertainty.
