India’s markets regulator is considering a domestically governed artificial intelligence model for the securities market, Chairman Tuhin Kanta Pandey said on 28 October. The announcement places AI development alongside a separate task: supervising how powerful automated tools are used by financial firms.
Reuters reported that the Securities and Exchange Board of India is working to implement the International Organization of Securities Commissions’ AI supervisory toolkit. Pandey described a possible market-specific domestic model as a form of digital public infrastructure. The remarks indicate exploration, not the launch of a publicly available trading product.
What has and has not been announced
The Reuters account identifies investor services, risk assessment and supervision as areas where AI could help. It also records Pandey’s emphasis on reliable data, model oversight and accountability for decisions.
Those qualifications are central to the proposal. A tool that spots an unusual pattern is not necessarily capable of explaining why it appeared, establishing wrongdoing or deciding an appropriate response. Different uses require different standards of evidence and control.
The sources reviewed do not set out a model specification, procurement award, deployment timetable or final budget. This report therefore does not describe the proposed system as operational, mandatory for brokers or available to retail investors.
The case for domestic governance
A securities-market system governed domestically could, in principle, offer closer control over sensitive information and the rules governing access. It could also be designed around local market structures and regulatory requirements.
The Free Press Journal reported that the proposal is being discussed against concerns about the confidentiality and availability of advanced models. That is an institutional rationale for exploration, rather than proof that a domestic system would automatically outperform alternatives.
Domestic ownership and effective governance are separate questions. Whoever builds the system must still demonstrate security, reliability and suitable controls. A model’s location does not by itself establish the quality of its training data, the fairness of its outputs or the independence of the people reviewing them.
An international toolkit, not a universal licence
IOSCO’s published toolkit is intended to help regulators approach AI oversight proportionately. The report describes practical, non-binding supervisory tools that can be adapted to different markets and regulatory systems.
Its purpose is to support supervision while maintaining investor protection, market integrity and financial stability. That framework does not remove a regulator’s responsibility to decide how a particular deployment should be tested or constrained.
VONG consulted an accessible copy of the IOSCO report because direct retrieval of the official PDF was blocked during this check. The official document link is provided alongside that copy. The report is background on supervision, not evidence that every measure it discusses has already been adopted in India.
Benefits for firms and investors
For regulated businesses, a well-designed tool could help organise complex records, identify matters requiring review and improve consistency in routine processes. For supervisors, it could help direct limited attention towards areas where more investigation is warranted.
For investors, the relevant benefit is not the sophistication of the model but the quality of service and protection it enables. A faster response is useful only if it is accurate and if errors can be challenged.
The chairman’s recorded NDTV Profit appearance also placed technology within a wider market-access and investor-understanding discussion. That context matters: making a system easier to use does not necessarily make the financial product behind it easier to understand.
The risks of automated authority
A model can produce a confident answer from incomplete or outdated information. If staff treat that confidence as evidence, an assistance tool can acquire more authority than its reliability warrants.
A hypothetical false alert illustrates the problem. A legitimate transaction might be marked unusual because it differs from the system’s training examples. The alert could still be useful, but a human investigator would need the underlying records before drawing conclusions.
This is an explanatory example, not an allegation about SEBI or an identified firm. It shows why validation, access controls and a documented review process matter as much as model capability.
Accountability, outsourcing and operational resilience
The Free Press Journal also reported work on a technology roadmap and proposed cybersecurity oversight involving market infrastructure institutions and their subsidiaries. These are related institutional measures, not features already demonstrated by the prospective AI model.
A financial system can depend on several suppliers even when its main model is domestic. Computing infrastructure, software updates, data feeds and specialist services can introduce separate dependencies.
Operational planning therefore needs to address more than a benchmark score. Institutions need to know what happens when a service fails, when data is corrected or when an update changes behaviour. Responsibility should remain identifiable across that chain rather than disappearing into a vendor relationship.
The next evidence to watch
A detailed proposal would allow the public to assess who controls the system, which data it can access, how it is tested and whether affected people have a route to challenge errors. A deployment decision would require its own evidence.
For now, SEBI has signalled interest in developing domestic AI capability while strengthening oversight. The opportunity is improved services and supervision. The unresolved test is whether any eventual system can deliver those benefits with transparent responsibility, dependable operation and safeguards proportionate to its role.
More coverage: VONG Technology.
