With AI/ML use growing across exchanges, brokers, and mutual funds, Sebi aims to strike a balance between innovation and regulation. The proposed framework covers governance, data security, bias prevention, investor protection, and risk controls.
The last date to send comments/suggestions is July 11.
Earlier, Sebi constituted a working group to study Indian, global best practices in AI/ML, mandating it to prepare guidelines for the usage of AI/ML applications. The working group was also tasked with providing recommendations to address concerns and issues related to the use of AI/ML applications.
Also Read: Sebi eases delisting norms for PSUs with over 90% government holding
Here are 5 guiding principles:
1) Model governanceMarket participants must set up internal teams with the technical expertise to monitor AI/ML models, document model development, and handle exceptions. Senior management will be held accountable for the entire AI lifecycle, including oversight of third-party vendors.
2) Mandatory disclosure
If AI/ML tools directly impact investors—like in algo trading or advisory services—firms must clearly disclose their use, including model purpose, risks, accuracy, and limitations. Language must be simple and client-friendly.
3) Robust testing & monitoring norms
Sebi proposes rigorous model testing in simulated environments before live deployment, and ongoing monitoring thereafter. Firms must retain data logs and documentation for a minimum of five years to ensure explainability and traceability.
4) Fairness and bias
To prevent discrimination, Sebi suggests using diverse, high-quality datasets and training staff to identify bias.
5) Data security
Companies must follow strong data governance, privacy, and cybersecurity protocols.
Also Read: Sebi board meeting: Regulator approves PSU delisting, IPO reforms, dematerialisation of Securities. 10 key takeaways
Sebi has suggested a lighter regulatory approach for internal-use models (e.g., surveillance), while models affecting clients would face stricter controls.
(Disclaimer: Recommendations, suggestions, views and opinions given by the experts are their own. These do not represent the views of Economic Times)
