Add us as a preferred source
The global head of life sciences at Tata Consultancy Services (TCS) explains that becoming AI-native does not mean relinquishing internal governance.
Artificial intelligence is advancing at a pace that outstrips traditional regulatory frameworks, leaving life sciences leaders to navigate an increasingly complex global landscape. Following the introduction of landmark legislation such as the European Union’s AI Act, pharmaceutical companies face stricter compliance demands than ever before. Yet as the industry evaluates how to safely and effectively harness advanced technologies, the question remains: how to accelerate innovation without compromising patient safety.
For Sanjeev Sachdeva, global head of life sciences at Tata Consultancy Services (TCS), meeting this challenge requires a shift in how the sector conceptualises technology. Rather than treating artificial intelligence as an external overlay, he advocates for the rise of the AI-native pharmaceutical organisation. It is one where automated tools and intelligence capabilities are woven directly into the fabric of enterprise workflows.
Here, he talks to Healthcare Today about why human oversight remains indispensable in pharmacovigilance, how the industry is guarding against model drift and data bias, and why proactive regulatory adoption is giving life sciences leaders the confidence to scale safely.
AI is transforming the pharmaceutical industry, but how do we ensure that the drive for innovation never comes at the expense of patient safety?
Patient safety is the fundamental principle that drives both pharma companies and regulatory authorities alike. Whether you are discovering new compounds or conducting clinical trials, the entire operational framework is designed to ensure patient safety remains the top priority.
As we look at the integration of AI, a common concern is the perception that these systems act as a black box capable of making autonomous choices. However, in critical domains such as pharmacovigilance and drug safety, the core principle remains unchanged: technology offers efficiency and insight, but accountability is never transferred to a machine.
AI provides extraordinary power – the power to automate workflows, drive operational efficiency, and, crucially for patient safety, detect early signals of adverse events. It can rapidly process, scan and identify complex clinical patterns across vast datasets in ways that would be humanly impossible for a person to undertake alone. In this sense, AI acts as a powerful intelligence assistant, augmenting human expertise while keeping clinical judgment firmly in human hands.
“Human intervention is not needed for every repetitive administrative action.”
How do you keep the human in the loop?
At a macro level, it is important to recognise that accelerating time-to-market does not mean every step requires human intervention. Rather, it is the critical, risk-sensitive decisions where a human in the loop is indispensable.
If a task simply involves extracting data and presenting it, agentic AI can handle the process end-to-end with no direct human involvement. However, when making a clinical judgment or validation, human oversight must be invoked to evaluate the information and make the final call.
Consider two contrasting examples. The first involves drafting complex regulatory submissions that require processing massive datasets alongside deep medical knowledge – such as preparing a benefit-risk assessment for a drug profile. Compiling these comprehensive aggregate safety documents traditionally demands a significant amount of time. AI can accelerate this by analysing trends under guided parameters and generating a comprehensive draft report for a medical expert to review, validate and sign off.
By contrast, when a healthcare professional initially reports an adverse event, the routine entry of that raw information into a safety database does not necessarily require human intervention. Because the initial step is purely operational case processing without clinical inferencing, AI can handle the entry directly.
Human intervention is not needed for every repetitive administrative action; it is deployed intentionally at critical decision points to ensure no additional risk is introduced to patient safety.
AI systems are only as good as the data on which they are trained. How can pharma companies guard against biased, incomplete or poor-quality data?
Data quality and bias represent critical challenges in AI implementation. If you train an agent or model on an unrepresentative dataset, the AI will accept that data as absolute truth and draw its correlations and inferences accordingly.
If a model is not trained on clinical data representing specific demographics – such as female trial cohorts or diverse regional populations – it simply cannot generate valid clinical inferences for those unrepresented groups.
To prevent these pitfalls, pharma companies must establish strict operational guardrails. Crucially, we do not simply feed raw data to an agent and allow it to learn unguided. Instead, we provide structured frameworks and explicit guidance on what the model can and cannot do. Training an AI agent is a curated, controlled process rather than an open-ended data collection exercise.
While no approach offers a 100% fail-safe guarantee against bias, risks are actively mitigated across two distinct operational phases. First, rigorous controls are applied during the initial training and prompt design stage. Second, once the agent is live in production, real-time monitoring and governance mechanisms continually evaluate its output.
Regulation is struggling to keep pace with technological development. How can pharma companies navigate evolving rules without allowing compliance to become either a tick-box exercise or a barrier?
Navigating regulation is perhaps the most complex challenge facing the sector. Modern pharmaceutical companies do not operate under a single national compliance framework; they navigate a global web across dozens of health authorities and local regulatory regimes, each with its own specific rules.
It is important to recognise that these compliance demands existed long before AI entered the picture. AI simply adds another layer of complexity to an already intricate system. Regulatory bodies are working to keep pace, issuing evolving frameworks and guidance rather than static, daily mandates. Industry groups are actively collaborating to establish baseline security and adoption standards.
This regulatory environment explains why the pharmaceutical sector is often perceived as a slow adopter of AI. It is not a lack of technical capability or financial capital; rather, it is a deliberate, risk-averse posture. Pharma companies want absolute certainty that they remain fully compliant with all published guidance – even when that guidance consists of broad or ambiguous frameworks that require extensive testing to interpret.
However, a very encouraging trend is taking shape. Major international regulatory bodies are not merely publishing AI guidelines – they are actively integrating AI into their own operations. A prime example is the FDA’s Sentinel Initiative, which relies on AI and advanced analytics to execute proactive, population-level drug safety surveillance.
When regulators themselves rely on AI to achieve their core safety missions, it gives the broader life sciences ecosystem the confidence to adopt these technologies. Companies can move forward in a measured, step-by-step fashion – navigating complex data security, personally identifiable information (PII) standards, and GDPR requirements alongside traditional health authority regulations without allowing compliance to stall innovation.
“Crucially, becoming AI-native does not mean relinquishing internal governance, clinical controls, or the core mechanisms that safeguard patient safety.”
Many companies are experimenting with AI, but moving from pilot projects to widespread adoption is much harder. What are the biggest obstacles?
Over the past two years – particularly as generative AI became a board-level conversation – the pharmaceutical industry was predominantly in an exploratory, pilot phase. However, as organisational confidence has grown, companies have begun actively scaling these deployments. While the sector has not yet reached a state of fully automated nirvana, the past 24 months have seen a decisive shift toward real-world deployment across the pharma value chain.
To build momentum, pharmaceutical companies initially targeted low-hanging fruit where compliance risk was negligible. They started with internal administrative functions like finance and HR, alongside knowledge-management systems designed to streamline internal documentation. Because these initial applications did not make critical clinical decisions, they provided a safe proving ground for testing capability.
Moving beyond these early wins from pilot-stage excitement into enterprise scale brings several practical considerations into sharper focus. Leadership now demands clear business value and demonstrable return on investment, which in turn requires clean, structured, and securely integrated enterprise data foundations. At the same time, companies must ensure model reliability and regulatory compliance, keeping outputs consistent, audit-ready, and fully compliant with global health standards while safeguarding proprietary data and patient privacy.
Finally, effective change management remains essential to shifting internal corporate culture and embedding new digital workflows into everyday workforce habits. Rather than stalling progress, these structured requirements are encouraging a measured approach to enterprise enablement, ensuring that scaling AI delivers lasting, safe business value.
Are we ever going to see AI design pharmaceuticals?
Yes, absolutely! To me, the ultimate destination is clear, though it will not involve 100% full automation. At TCS, we position this future around the concept of the AI-native pharmaceutical organisation. AI will not be an artificial layer imposed upon existing operations; instead, it will exist as an inherent, foundational component woven directly into every enterprise capability.
Crucially, becoming AI-native does not mean relinquishing internal governance, clinical controls, or the core mechanisms that safeguard patient safety. While we may never see a completely autonomous pharmaceutical company, the sector will transition from today’s narrow automation toward a state of high operational autonomy.
Looking ahead to 2030, a pharmaceutical organisation might not achieve 100% total autonomy, but reaching 70% to 80% autonomy is entirely realistic. That level of operational independence will be complemented by qualified clinical, medical and scientific experts ensuring oversight across research, regulatory compliance, data security and patient wellbeing. That hybrid model – combining high system autonomy with expert human oversight – is what will define a truly AI-native pharmaceutical enterprise by 2030.
