New Delhi: Imagine walking into a hospital where robotic footsteps echo through the halls.
A machine scans patients, recommends treatments, and updates records, all without waiting for a doctor’s command.
It feels almost too real, yet this is the promise of agentic AI: speed, precision, and relief for India’s overstretched healthcare system. But with great autonomy comes risk, as every automated decision carries the shadow of cyber threats and patient harm.
As we all know Artificial Intelligence (AI) has already begun reshaping healthcare, from predictive diagnostics to administrative efficiency. But a new frontier, agentic AI, is sparking both excitement and anxiety. Unlike traditional AI that follows instructions, agentic AI can act independently, adapt to new information, and make decisions in real time.
In India, where one doctor serves an average of 1,500 people, the promise is enormous — faster diagnoses, standardised care, and better access in underserved regions. Yet this potential is matched by vulnerabilities, with millions of cyberattacks targeting healthcare systems.
At a recent ETHealthworld webinar titled Healthcare on Autopilot: Understanding the Rise of Agentic AI, leading clinicians and digital health experts including Rajiv Sikka, Group CIO, Medanta Hospitals; Dr Rahul Bhargava, Principal Director and Head of Hematology, Hemato Oncology and BMT, Fortis Memorial Research Institute, Gurugram; Dr Sujoy Kar, Chief Medical Information Officer and Vice President, Apollo Hospitals; and Dr Sushil Kumar Meher, Head of Health IT and CISO, AIIMS Delhi, came together to weigh the opportunities against the challenges.
Informing that Agentic AI is no longer just theory, hospitals in India are already putting it to work, Sikka, shared how Medanta has deployed agentic voice AI in outpatient departments.
“During patient–doctor consultations, the AI generates a ready-to-use prescription. This isn’t just voice-to-text — it’s voice-to-prescription,” he explained. The system produces prescriptions in nine languages, reducing errors caused by illegible handwriting and improving accessibility. “The outcome? Doctors feel comfortable, patients feel heard, and the consult is more efficient,” he said.
Dr Rahul Bhargava, Principal Director of Hematology at Fortis Memorial Research Institute, comparing agentic AI is like autopilot in aviation,said,”IT doesn’t replace the pilot but makes the journey safer.”
He pointed to an app developed for bone marrow transplant patients that tracks symptoms for six months after surgery. “The real power is in standardization. With agentic AI, evidence-based practices can be applied whether in Delhi, Ludhiana, or Bareilly.”
But the technology is only as good as the data behind it, cautioned Dr. Meher.
“Garbage in, garbage out. Cleaning one dataset for one AI tool took three years and 40 doctors. Without quality data, AI is a black box. And who certifies the accuracy of these models before deployment? In healthcare, 65% accuracy is not acceptable,” he said.
Adding that scientific validation is lacking, he said, “Models should be tested on 4,000–5,000 samples. That’s not happening.”
Meanwhile Sikka pointed to progress under the Ayushman Bharat Digital Mission (ABDM) and the creation of ABHA IDs and said, “For the first time, healthcare policy is embedding digital standards like ICD-10, HL7, and FHIR. This will unlock interoperability and allow AI models to scale.”
Dr. Sujay Kar, CMIO & VP, Apollo Hospitals, underlining the need for rigor, said, “Enthusiasm must be matched with safeguards. We focus on three pillars: data quality, NLP and process design, and clinical accuracy. Without these, deployment is unsafe.”
He warned of risks like adversarial prompts that could alter prescriptions, said “turning 500 mg into 5,000 mg of paracetamol and hallucinations that still persist. Apollo uses a “human-in-the-loop” approach. AI can automate low-risk tasks like scheduling, but human oversight is mandatory in high-risk clinical scenarios.”
Who Is Accountable?
If an AI model makes a wrong medical recommendation, who bears responsibility? According to Dr. Meher, under Indian law, the patient owns their data, while the hospital and doctor are custodians. “If something goes wrong, the liability lies with the hospital and physician, not the AI vendor.”
This lack of clarity underscores the urgent need for independent validation and regulation. The European Union is moving ahead with its AI Act, but India has yet to establish healthcare-specific frameworks.
The panel agreed that agentic AI is already showing promise in easing clinical workloads, improving standardisation, and expanding access. Yet the risks like poor data quality, lack of independent validation, cybersecurity threats, and regulatory grey zones cannot be ignored.
As per the expert panelist India needs to work on to build India-trained, specialty-specific AI models. Mandate independent validation and certification of medical AI. Use “human-in-the-loop” systems for high-risk decisions. Accelerate ABDM adoption to enable standardized, interoperable data.
Agentic AI could be India’s most powerful tool for scaling healthcare access or its most dangerous experiment. As Dr. Bhargava put it “AI augments doctors; it doesn’t replace them. But without ethical safeguards and data security, the risks could outweigh the rewards.”
Yet the debate remains open if agentic AI healthcare’s biggest opportunity or its biggest risk?
