IFSCA · Circular

Advisory on Heightened Cyber Security Risks arising from Frontier Artificial Intelligence Models

IFSCA-CSD/MSC/3/2026-DCS

In short
IFSCA-CSD/MSC/3/2026-DCS is the Advisory on Heightened Cyber Security Risks arising from Frontier Artificial Intelligence Models, in force from 4 June 2026. It is called an advisory and its covering circular says regulated entities are encouraged to comply — but six of the eleven items in Annexure A are drafted with "shall", and two more wrap a shall limb inside an encouragement. It appears to be the first Indian financial-sector instrument to name frontier AI as a defined scenario a board has to be shown.

The record

ReferenceIFSCA-CSD/MSC/3/2026-DCS
Issued byInternational Financial Services Centres Authority (IFSCA)
Instrument typeCircular
Date of issue4 June 2026
StatusIn force
BindsAll regulated entities in the IFSCs, read alongside the 2025 Guidelines and, for an MII, the 2026 MII Guidelines — it expressly dilutes neither. Titled an advisory, but six of the eleven Annexure A items are drafted with “shall”: SBOM coverage, an API inventory with rate-limiting, frontier AI as a named board-level risk scenario, and critical-service-provider assurance among them.
Dates it sets
  • 4 June 2026In force with immediate effect.
IFSCA Frontier AI Cyber Advisory, 2026 as the register holds it, read from the issuing regulator's own notification.

What it says

  • Requires regulated entities to presume newly disclosed critical vulnerabilities are exploitable within hours.
  • Requires frontier AI to be carried as a named scenario in the cyber risk assessment, reviewed periodically and put before the Board — before the Standing Committee on Technology at an MII.
  • Requires a Software Bill of Materials covering open-source components, and a comprehensive inventory of APIs and the applications consuming them, with rate-limiting and a whitelist.
  • Requires critical service providers to assess frontier-AI risk and furnish evidence of preparedness.
  • Requires monitoring tuned for attack sequences that exceed plausible human-operated timelines, and human oversight plus security testing of AI-generated code before production.

What accounts of this instrument get wrong

Read the instrument

The link below goes to the issuing regulator’s own copy. Where a regulator serves its text through a PDF viewer or a query-string URL, it points at the document rather than at a landing page that may not render it.

Not a reading list. Each of these alters the obligation on this page — the amendment that moved a date, the sibling that binds the entity class this one excludes, or the separate track it is routinely merged with.

This record is one entry in the Indian cyber regulation register, which publishes the same fields for every instrument as an open dataset. Indicative, and not legal advice: whether an instrument reaches your organisation turns on your licences and registrations, and is a determination for your legal team.

Every instrument cited here was verified against the issuing regulator's own notification on .

Knowing the instrument is not knowing where you stand.

Every instrument on this page asks an organisation to know its own exposure. Your organisation already has a security rating, calculated from signals anyone outside it can see — including whoever supervises you. Request the complimentary Cyber Risk Rating Report and read what those signals say.

Request my rating Who is accountable when an AI agent causes the incident