An AI answer can sound certain while using the wrong date, the wrong denominator or a source that never supported its conclusion. That is why the discussion around SEBI’s planned AI safeguards matters to ordinary investors, even if they never read a model-governance policy.

A PTI report dated 15 November 2026 describes the SEBI chairman’s remarks about forthcoming detail on responsible AI use, including accountability, disclosure and human oversight. Those remarks should not be mistaken for the text or commencement date of a new circular. Existing requirements also do not disappear while further guidance is discussed.

Source: PTI report, 15 November 2026, on the SEBI chairman’s AI-guidance remarks.

SEBI’s June 2025 consultation is a separate, identifiable document. It discusses governance, investor protection, testing, fairness, privacy and security. The checklist below is our practical reading discipline for investors, rather than a reproduction of draft obligations or a claim that every item is now a statutory rule.

Source: SEBI AI/ML consultation paper, 20 June 2025, PDF pages 2–3 and 7–13; proposals, not the November announcement.

CheckAsk the toolA useful answer should provide
1. SourceWhere is the original evidence?A relevant document or data record
2. DateWhat period does this describe?Event, observation and publication dates
3. CalculationCan I reproduce the number?Inputs, formula, units and assumptions
4. ComparisonAre these like-for-like figures?Consistent period, category and denominator
5. UncertaintyWhat is missing or estimated?Explicit limits and alternative explanations
6. PrivacyWhat am I being asked to upload?A clear purpose and data-handling explanation
7. AccountabilityWho can review a consequential error?An identifiable support or review route

1. Open the source, not just the citation badge

A link is useful only if its contents support the claim. Open the document and look for the relevant statement or table. A general homepage, a topic index or an older report can look credible without substantiating a current number. The check is relevance, not merely whether a familiar institution’s name appears beside the answer.

For example, a report about last year’s bond operation cannot establish a new operation this week. The explanation of bond-price mechanics might still be sensible; the event claim would remain unsupported. Keeping those two parts separate lets you retain useful analysis while rejecting a false news hook.

2. Match the date to the question

An article published today may describe yesterday’s close, last month’s holdings or an older regulatory change. Ask for the observation date as well as the publication date. A portfolio snapshot is particularly easy to misread as a live position. A useful response should tell you how fresh the inputs are and where more recent evidence would be needed.

This is also important for rules. A consultation date, a final notification date and an effective date answer different questions. A model that compresses them into one date can give the right-looking rule at the wrong point in time.

3. Recalculate a small example before trusting a large one

Consider a hypothetical investment starting at 100. It falls 10%, reaching 90, then rises 10%, reaching 99. Adding the percentages to claim a flat result would be wrong. The correct calculation is 100 × 0.90 × 1.10. Nothing about fluent prose changes that arithmetic.

For a financial calculator, ask whether contributions, withdrawals, dates, expenses and taxes are included. A lump-sum return and a cash-flow-based return do not answer the same question. If a tool cannot state the inputs and timing, a very precise output should be treated as a reason to investigate rather than a reason to trust it.

Illustration: equal percentage moves do not cancel

Index, starting at 100 · Illustrative data

LabelValue (Index, starting at 100)
Start100
After a 10% fall90
After a subsequent 10% rise99
Fictional investment path: 100 × 0.90 × 1.10 = 99. Not market history or a return forecast. Source: MF Analyser illustrative arithmetic

4. Check the denominator before comparing funds

A stock can be a small share of the whole portfolio and a larger share of its equity portion. Both percentages may be correct. Without the denominator, however, two valid numbers can appear to contradict each other. Ask whether the calculation refers to total invested value, disclosed equity, one scheme or a selected subset.

The same discipline applies to overlapping funds. Counting common stock names is different from measuring common portfolio weight. An answer that merges those concepts can exaggerate or understate diversification. A strong tool explains which measure it calculated and how missing holdings affect the result.

5. Make uncertainty visible

An incomplete transaction history does not support a confident personalised return. An unidentified tax lot does not support a precise tax bill. A future market scenario does not become an observed probability because an AI assigns a percentage. Ask which parts are reported facts, calculated outputs, assumptions and open questions.

Good uncertainty language should lead to an action: supply the missing transaction, check the relevant document or use a range under stated assumptions. A vague warning under a confident answer is less useful than identifying the exact input that could change the conclusion.

6. Share only what the task needs

Before uploading statements, identify who operates the service, why the information is needed and how it is handled. A general question about a fund’s published holdings should not require unrelated personal documents. A personalised analysis may need transaction data, but that still deserves a clear explanation of retention, access and deletion arrangements.

Avoid placing passwords, one-time codes or payment credentials into a financial chatbot. Where a legitimate workflow needs identity or account access, use its designated secure process. The convenience of a chat window does not remove the need to understand the destination of your information.

7. Know who reviews the difficult decisions

Use AI to organise research, explain a formula and identify questions worth asking. Before a consequential financial action, make sure the reasoning is traceable and the relevant professional or regulated service can address suitability and legal requirements where applicable. A model’s confidence is not a substitute for accountability.

Save the answer, its date and the evidence behind any important decision. If an input changes, you can then test whether the conclusion should change too. That is a stronger standard than asking whether an answer sounded impressive: can it be checked, challenged and corrected?

Questions readers ask

Do the November AI remarks establish a new compliance deadline?

The cited remarks describe further guidance. They do not themselves supply the final regulatory text or an implementation deadline.

How can I check an AI investment calculation?

Request the inputs, units, dates, formula and assumptions, then reproduce a small example independently.

Is a citation enough to establish accuracy?

No. The linked document must actually support the claim and match its observation period and scope.

Put the analysis beside your own investments

When using Fund Research or Portfolio Analysis, keep the data date and coverage beside the result. This checklist applies to MF Analyser as well as to any other tool; branding does not exempt an answer from verification.

Research notes: Checked on 19 November 2026 against the dated documents linked below. Calculations state their inputs; illustrative scenarios are not forecasts. Cover: generated editorial illustration, not a photograph documenting the event. Educational analysis; no personalised investment recommendation.