SEBI AI Disclosure Rules for Research Analysts (2026)
Almost every Research Analyst uses AI now. SEBI knows that, and the rule is simple: tell your clients where you use it, and own what comes out. You cannot hide behind the model. Here is what the disclosure has to cover, what to write, and how to use AI without handing your accountability to a black box.
Note: General information for Research Analysts, not legal advice. Confirm the current wording against the SEBI Regulations and the latest FAQs before you publish.
What the rule says
Regulation 19 of the SEBI (Research Analysts) Regulations requires you to disclose the extent to which AI tools are used in providing research services. Two ideas sit behind it. First, transparency: a client has a right to know whether a human or a model wrote the analysis they are paying for. Second, accountability: using AI does not move the responsibility off you. If the output is wrong, that is your problem, not the vendor's.
Write the disclosure with scope, not vibes
A line that says โwe may use AIโ tells a client nothing and tells an inspector even less. Be specific. State which part of the work the tool does, and where the human takes over. A clean version reads like this:
โWe use AI tools to summarise public filings and news and to draft the factual portion of our notes. A SEBI-registered Research Analyst reviews, edits and approves every note before it is sent. The analyst makes the recommendation and remains responsible for it. No client data is shared with the AI tool; only public filing and news text is processed.โ
That covers scope, the human-in-the-loop, accountability, and data handling in four sentences. Put it in your terms and your report disclosures.
The line you cannot cross
AI can write the boring eighty percent: the summary of a results filing, the key numbers, the standard structure. What it cannot do is make the call. The recommendation, the view, the judgement that you are registered to provide, has to come from you. Treat every draft as a first pass to be checked against the source, not as finished research. An auditor who sees unedited model output sent straight to clients will treat it as a finding.
Client data never goes into the model
SEBI also holds you responsible for client-data security when you use these tools. The safe rule is blunt: the model sees the public filing, never your client book. Do not paste client names, holdings, PAN or contact details into a general AI tool. If your workflow keeps the two separate by design, your AI-use disclosure is honest and your data risk is low.
This is exactly how Aktai is built. The AI reads the BSE or NSE filing and drafts a factual note in seconds. It never sees your clients. You edit, add your view and send under your own name, and every send is logged in your Regulation 25 audit trail. The disclosure writes itself because you know precisely what the tool did and did not do.
FAQ
Does SEBI require Research Analysts to disclose AI use?
Yes. Under Regulation 19 of the SEBI (Research Analysts) Regulations, you must disclose the extent to which AI tools are used in providing research services. The disclosure goes to clients, and you stay fully responsible for the output and for the security of any client data the tools touch.
What should the AI disclosure actually say?
Keep it plain and specific: which part of the work the tool does (for example, drafting a factual summary of a filing), that a registered analyst reviews and approves every output before it goes to a client, and that the analyst remains responsible for the recommendation. Avoid vague lines like "we use AI" with no scope.
Can AI make the recommendation for me?
No. The model can draft the factual base and save you time, but the recommendation and the responsibility stay with the registered analyst. You review, edit, add your view and sign. That separation is the whole point of the accountability rule.
Does using AI put client data at risk under SEBI rules?
It can if you are careless. You remain responsible for client-data security regardless of the tool. The safe pattern is to send the model only the public filing or article text and tickers, never client names, holdings or contact details. Aktai is built this way: the AI sees the filing, not your client book.