As governments across the world are adopting artificial intelligence (AI) to speed up service delivery, recent instances of AI agents acting autonomously have sparked debates over how much responsibility should be delegated to increasingly capable AI systems.
From e-governance tools to generative AI, governments have used new technologies to digitise services, support decision-making, assist officials, and make public services more convenient for citizens. The shift to agentic AI, however, raises concerns regarding what happens when AI moves from assisting a government official to performing parts of the administrative process itself.
The issue, therefore, is no longer limited to whether an AI gives the correct answer, but what it should be authorised to do, and who remains accountable for its decisions.
Generative AI in governance
Generative AI primarily provides information, summaries and creates new content such as images and videos based on prompts by users, while agentic AI has the ability to use digital tools, access information and undertake a sequence of actions to achieve a defined objective, thereby acting as an agent with considerable autonomy.
Conventional AI has been integrated into governance for quite some time. In healthcare, AI has enabled faster disease diagnosis, drug development, improved treatment recommendations and data-driven decision-making. In agriculture, AI is already being used for pest surveillance, crop monitoring and weather-related advisories. The National Pest Surveillance System uses AI and machine learning to detect pest infestations and support timely intervention.
AI has also contributed to transforming governance and the delivery of public services. For instance, Kisan e-Mitra, a voice-enabled, AI-powered chatbot developed by the Ministry of Agriculture and Farmers’ Welfare, is an example of the use of AI in public service delivery.
Story continues below this ad
The chatbot assists farmers by answering questions about major government agricultural schemes, such as PM-KISAN (Pradhan Mantri Kisan Samman Nidhi), PMFBY (Pradhan Mantri Fasal Bima Yojana), and KCC (Kisan Credit Card). The platform operates in 11 regional languages and handles more than 8,000 farmer queries a day. As of March 2026, it has answered more than 95 lakh queries.
E-Mitra exemplifies how digital tools help improve operational efficiency, expand institutional outreach, and enable faster access to services and information.
Agentic AI: from assistance to action
But agentic AI is not only able to do more than generative AI, but also can have a greater degree of autonomy and access. Agentic AI is a model that executes tasks for users without constant prompts, but within certain limits. It can pursue a defined goal, going through multiple steps, using various tools and interacting with its environment.
If designed to do so, AI agents can gather information, formulate a plan, access databases or applications, and execute tasks, with different degrees of autonomy. Adding an agentic component to an existing AI-enabled service could take it beyond providing information and assistance.
Story continues below this ad
In agriculture, for instance, the use of agentic systems could potentially integrate crop data, advisories, and government schemes to provide more personalised and accurate assistance to farmers. In healthcare, it could assist with administrative tasks, such as appointment scheduling, follow-ups, documentation of records, and coordination between various medical institutions and healthcare services. It can also be used in the coordination of emergency response.
Similarly, in the sphere of broader administrative mechanisms across sectors, AI agents could support functions like document verification, grievance processing, and coordination between various departments. This could reduce the routine workload of officials and allow governments to handle large volumes of applications and requests, thereby reducing waiting periods for citizens. The ability of AI agents to operate around the clock and provide services in multiple languages could potentially improve public services in a multilingual country like India.
Institutionalisation of AI in governance
Therefore, governments across the world are exploring the ability of agentic AI. Earlier this year, Singapore launched its Model AI Governance Framework for Agentic AI. In April 2026, the UAE announced its plans to shift 50 per cent of its government services and operations to agentic AI within two years.
Abu Dhabi also renamed its Ministerial Development Council as the Ministerial Council for Artificial Intelligence and Development and charged it with the task of integrating AI into government work. The initiative is an important shift from using AI primarily for analysis and assistance, to being used as a system that is able to plan and execute tasks.
Story continues below this ad
Notably, countries are also appointing dedicated AI ministers. The UAE has minister for AI since 2017. In July 2026, the UK appointed its first dedicated AI minister, Kanishka Narayan. In India, the Keralam government has created a dedicated AI portfolio for the first time at the cabinet level. Tamil Nadu has also appointed R Kumar as Minister for AI, IT and Digital Services.
Is AI simply a technology portfolio?
But these developments raise a key question: whether AI is simply a technology portfolio or is it becoming embedded in the function of a state? It may be illustrated with the example of Kisan e-Mitra. The AI-powered app helps farmers overcome the need to identify useful schemes and to get help from officials or service centres to apply for them.
But adding an agentic component to the app could potentially go further. It could access farmer records, and prepare, submit and track applications. Each of these steps requires a new level of administrative authority. While accessing records involves handling personal data, submitting applications constitutes an administrative action.
But as more stages could be delegated to agentic AI, the potential ‘blast radius’ of an error, which is the number of people or transactions affected by a single error, also increases. Therefore, while the use of agentic AI in governance has the potential for significant positive transformation, it also carries significant risks.
Story continues below this ad
Risks and challenges
First, AI is prone to errors and hallucinations. Generative AI systems can produce incorrect information with confidence. When an AI agent is authorised to act on such information, it could lead to incorrect application or administrative action, with potential consequences for citizens. Singapore’s Model AI Governance Framework highlights risks arising from agents taking unauthorised or erroneous actions.
Second, agentic AI creates a complex question of accountability. An erroneous or unauthorised action by agentic AI could be due to reasons like the underlying AI model, the data on which it relies, or the workflow process. But an AI system cannot itself bear legal or administrative responsibility, which underlines the significance of establishing clear lines of responsibility.
Third, bias and discrimination could be seen as yet another challenge. AI systems could potentially reproduce existing patterns present in data on which they are trained. Fourth, there are concerns over privacy and cybersecurity, as AI agents with access to several databases and applications increase the risk of data breach, misuse, or unauthorised access. Agentic systems are also vulnerable to threats such as prompt injection and agent hijacking, wherein malicious instructions from external content can influence AI behaviour.
Fifth, the digital divide remains yet another concern, particularly in a country like India. If government services increasingly depend on AI-enabled interfaces, citizens with limited digital literacy, connectivity and access to digital services could face new barriers. AI-enabled governance, therefore, cannot replace accessible human and offline channels altogether.
Story continues below this ad
Finally, the environmental costs of AI infrastructure are also a matter of serious concern. AI systems require data centres that need a significant amount of electricity and, in many cases, clean water for cooling. This may aggravate resource pressure.
Governing agentic AI
Therefore, it may be argued that the approach to agentic AI combines the benefits with well-defined limits on autonomy. While low-risk administrative processes may be allowed to have greater AI autonomy, decisions affecting welfare entitlements, health, employment and other important domains can have meaningful human oversight.
In addition, there is a need to establish a clearly identifiable human or institutional authority responsible for AI actions to ensure accountability. Government departments can maintain audit trails that show the information accessed, decisions made, and actions undertaken by an agent. Singapore’s AI framework stresses that humans remain ultimately accountable and recommends clearly defined points for human approval in significant decisions.
Moreover, the principle of least-privilege access can be applied to the agentic systems. An AI agent can have access only to the data, permissions and system necessary to perform its designated task. This may help limit the consequences of errors or cyberattacks.
Story continues below this ad
There should also be clearly defined points at which the AI agent must defer to a human official. This is particularly important in decisions involving discretion, conflicting information, or a significant impact on an individual’s rights or entitlements.
Need for safeguards
The evolution from generative to agentic AI is not only a change in technology but also in the nature of digital governance. In the context of India, an important task would be to determine which government functions can be delegated to AI, what permissions such systems can have, and where human oversight and decision-making can remain necessary.
The establishment of the AI Governance and Economic Group (AIGEG) under the Ministry of Electronics and Information Technology (MeitY) is a step in the right direction to build an institutional framework for AI governance. Ultimately, the effective use of agentic AI will depend on the quality of government data and the safeguards governing AI actions.
Post read questions
1. AI-enabled governance can improve administrative efficiency, but excessive reliance on autonomous systems may create new forms of exclusion and accountability gaps. Examine.
Story continues below this ad
2. Distinguish between generative AI and agentic AI. How does the latter change the nature of risks associated with the use of AI in public administration?
3. Can agentic AI improve the efficiency and accessibility of public service delivery? Explain with examples.
4. The transition from generative AI to agentic AI could fundamentally alter the way governments deliver public services. What safeguards are necessary before delegating administrative functions to AI systems?
5. Discuss the institutional and regulatory measures required to ensure accountable deployment of agentic AI in Indian governance.
(Kannan K is a Doctoral candidate at the Centre for Economic and Social Studies, Hyderabad.)
Share your thoughts and ideas on UPSC Special articles with ashiya.parveen@indianexpress.com.
Click Here to read the UPSC Essentials magazine for September 2026. Subscribe to our UPSC newsletter and stay updated with the news cues from the past week.
Stay updated with the latest UPSC articles by joining our Telegram channel – IndianExpress UPSC Hub, and follow us on Instagram and X.
Leave a Reply