What happened
SEBI published a consultation paper on the responsible use of AI and machine learning in securities markets in June 2025. In August 2026 its chair said final guidelines would follow shortly, and that they would require kill switches, human-in-the-loop controls and data controls. Regulated entities will remain responsible for AI tools whether they build them or buy them.
The consultation also proposed that firms disclose AI use to clients where it affects them, test models before and after deployment, and keep model documentation and input and output data for at least five years.
What it means for you
For brokers, asset managers, advisers and market infrastructure, the expected rules match how agents should be built anyway: limits on what the system can do, a person approving decisions that affect clients, and a way to stop it quickly. The five-year record keeping proposal matters most for agents, because it covers every input and output, not just the final decision.
What to do this quarter
- Inventory AI and ML tools used in trading, advice, surveillance, KYC and client service, including vendor tools.
- Confirm each has a tested off switch and a named person accountable for it.
- Check that your logging can keep inputs and outputs for five years.
Sources
This update is general information, not legal advice. Check how it applies to you with your legal team.
