The Securities and Exchange Board of India (SEBI) will soon issue detailed guidelines for the responsible use of artificial intelligence (AI) in the capital markets, SEBI Chairman Tuhin Kanta Pandey said at the India International Trade Fair in New Delhi on Saturday, PTI reported. PTI’s report does not specify a timeline.

According to PTI, Pandey said the guidelines will set guardrails for AI use, including:

  • Disclosures
  • Accountability of regulated entities
  • A “human-in-the-loop” approach: Human oversight over critical decisions.

He said SEBI is reviewing its regulatory framework to ensure that market participants’ use of AI does not compromise investor protection, market safety or integrity.

SEBI already holds firms accountable for the AI tools they use. In December 2024, it amended its regulations to make regulated entities solely responsible for the outputs of AI tools they use, whether they develop or procure them, as well as for the privacy and security of investors’ data handled by those tools.

What did SEBI propose in 2025?

SEBI released a consultation paper on the responsible use of AI and machine learning (ML) in securities markets on June 20, 2025, and sought comments until July 11, 2025. The paper builds on the recommendations of a SEBI working group, which called for a “human-in-the-loop throughout the development and deployment lifecycle.” The paper proposed that:

  • Disclosure to investors: Firms using AI in operations that may directly affect clients, such as algorithmic trading, portfolio management and advisory services, should inform those clients. Disclosures should cover the model’s features, purpose, risks, limitations and accuracy, any fees, and the quality of the data it uses, in language clients can understand.
  • Accountability for vendors: AI services provided by third-party vendors “are deemed to be provided by the market participants,” who remain responsible for complying with the law.
  • Senior oversight: Firms should designate senior management to oversee their AI models, establish fallback plans in case a model fails, and appoint a team independent of the model’s development to audit it. Firms would also have to share audit findings and model accuracy results with SEBI.
  • Testing: Firms should test models in an environment separate from live systems, conduct shadow tests using live traffic before deployment, and continue monitoring models after launch. They should retain input and output data for at least five years.
  • Fairness: AI models “should not favour or discriminate one group of clients/customers over another.”
  • Data privacy: Firms should comply with data protection laws and report data breaches and technical glitches to SEBI.

The paper proposed lighter requirements for internal uses of AI, such as compliance, surveillance and cybersecurity tools.

It also identified market-wide risks and suggested possible controls:

  • Deepfakes and misinformation: The paper flagged that generative AI could create fraudulent financial statements, misleading news reports and deepfakes that could manipulate market prices. It suggested watermarking AI-generated content.
  • Reliance on a few AI providers: It warned that several firms relying on a small number of generative AI providers could create systemic risks. It suggested that firms report the AI vendors they use and that dominant AI providers face closer monitoring as critical service providers.
  • Herding: It warned that firms using similar models and datasets could make similar market decisions, potentially moving markets in the same direction. It suggested circuit breakers and regular audits to detect collusive patterns.
  • Runaway AI: It flagged the risk that flaws in generative AI systems could spread across markets. It suggested stress testing, circuit breakers and kill switches.

Is a human in the loop enough?

Pandey has said that human oversight alone is not enough. Speaking at the Global Fintech Fest in September, he said that “humans can also overtrust machines” and that oversight “must be competent, empowered and accountable.” He added that “an AI-generated alert is not a finding.”

What else did Pandey say?

According to PTI:

  • SEBI’s AI tools: SEBI has developed several supervisory tools, including one called Sudarshan, which it is demonstrating at international regulatory forums. SEBI’s annual report spells the tool’s name as SUDARSAN and says it scans public social media content for investment fraud. The tool has flagged more than 20,000 instances of fraudulent content since it began operating in November 2025.
  • Cybersecurity: Pandey highlighted growing concerns over cyberattacks on financial institutions and cryptocurrency platforms, as well as digital fraud involving social media influencers. He said SEBI introduced a structured incident response mechanism in May, aligned with the Financial Stability Board’s reporting template. That month, SEBI also directed all regulated entities to overhaul their cybersecurity measures amid concerns over Anthropic’s Claude Mythos model.
  • Quantum computing: SEBI is taking an inventory of the areas in which markets use cryptography to assess their vulnerabilities to quantum computing. Pandey said markets expect these risks to become more pronounced between 2032 and 2035 and need to prepare well in advance.
  • Startups: SEBI has formed a working group to make the Innovators Growth Platform more attractive to startups, particularly deep-tech and biotech companies that take a long time to become profitable. Pandey said the platform has not gained the desired traction.

Also Read: