India deploys AI agents faster than it can govern them: the real risk isn't autonomy, but accountability
Jamil Khatri, CEO of Uniqus Consultech, raises a fundamental warning in Financial Express: India has the talent and infrastructure to lead the adoption of enterprise AI agents, but still lacks the governance framework to control the authority it cedes to them, permission by permission.
By Momentum IA · June 28, 2026.
There is a sentence in Jamil Khatri's article worth pausing on: «The risk does not begin when a machine starts running a company. It begins earlier, when decision-making authority is handed over one permission at a time». It is an uncomfortable observation because it describes something that is already happening, quietly and without headlines.
The piece, published as an op-ed in The Financial Express, describes the typical trajectory of an AI agent in finance: first it detects mismatched invoices, then it contacts suppliers, then it recommends adjustments and routes transactions for approval. A customer service agent goes from drafting responses to issuing refunds within certain limits. Each step seems reasonable; together they amount to a silent transfer of executive authority to a software system. What stands out is that this progression is not deliberately designed: it simply happens, use case by use case, as teams expand the system's permissions to gain efficiency.
Khatri argues that India has solid foundations to adopt these agents at scale and at speed. The country has repeatedly shown that it can compress technology costs without sacrificing reach: mobile data arrived at a fraction of the global price, digital payments approached a practically zero marginal cost per transaction. These are real and powerful precedents. However, and this is the core of the article, that capacity for fast and cheap adoption does not equate to operational readiness. Many business processes in India remain fragmented, poorly documented or dependent on tacit human judgment that has never been formalized. Automating a chaotic process does not order it: it perpetuates it, but now without a person to catch the anomaly.
The problem becomes more acute when agents enter finance, regulatory compliance or regulatory reporting. Once a system begins reconciling accounts or preparing reports for the regulator, it is no longer «alongside» the control environment: it is the control environment. And existing controls, designed on the assumption that a human initiates, reviews or approves each action, are left without a foundation.
Our read: this is not a problem unique to India, but India embodies it with particular urgency because the speed of adoption is going to be notable. The article identifies a governance gap that already exists in companies around the world and that will tend to widen the more capable and more autonomous the agent becomes. The question is not whether there will be failures —there will be— but who will bear responsibility when an agent makes a wrong decision with material consequences. Khatri is explicit: the company cannot shift that responsibility onto the model or SaaS platform provider. Outsourcing the software does not outsource accountability.
The five-question framework he proposes —where are the agents deployed?, what authority do they have?, who is responsible for each material outcome?, are the relevant actions logged?, can a human intervene or stop the system?— is simple but brutally practical. Most companies reading those questions will not be able to answer all five clearly. That in itself is a diagnosis.
As sector context, the tension between technical capability and governance maturity is not new: we lived through it with financial risk models before 2008, with content moderation algorithms on social media, with automated credit scoring systems. In each case, the industry ran faster than regulation and faster than its own internal controls. With AI agents the cycle repeats, but the speed of propagation is greater and the permissions ceded are broader.
In the long term, automating finance, compliance and operations tasks can free thousands of professionals from repetitive, low-value work to devote themselves to judgment, strategy and human relationships. That is the horizon worth defending. But getting there requires exactly what Khatri calls for: that the same discipline with which India scales technology be applied to process design, to authority limits and to accountability mechanisms. Without that prior work, agents' errors will not be technical anecdotes: they will be operational risk events with a first and last name on the balance sheet.