An Indian bank, insurer, hospital group or listed manufacturer that deploys AI today answers to several rulebooks at once. Its global customers and auditors increasingly ask about the NIST AI Risk Management Framework and ISO/IEC 42001. Its legal team is working through the Digital Personal Data Protection Rules, notified in November 2025, with most obligations due to apply after an 18- month window. Its sector regulator has published, or is consulting on, guidance of its own. Each instrument has its own vocabulary, and each arrives with its own consultants, templates and audit requests.
The result in many enterprises is four parallel compliance projects for one set of AI systems. That is expensive, and it is also risky, because controls designed separately leave gaps where they meet. A single AI risk management framework, built once and mapped to every instrument the enterprise must satisfy, is both cheaper and safer.
This article sets out what each of the main instruments actually asks, offers a unified control map that combines them into one set of controls, adds a risk-tiering rule so that low-risk tools are not buried under controls designed for credit decisions, and suggests an implementation sequence keyed to the DPDP timetable. Board-level AI governance is covered in a separate PanScience Research piece; this article sits one level below, at the controls that management operates.
Why one framework instead of four
The instruments overlap more than their branding suggests. Almost all of them ask the enterprise to know which AI systems it runs, to assess their risks before deployment, to assign accountable owners, to monitor systems once live, to respond to incidents and to tell affected people what is happening. They differ in emphasis, legal force and detail. If each is implemented on its own, the enterprise ends up with several AI inventories, several risk assessment templates and several incident processes, none of them complete.
A unified framework reverses the logic. The enterprise designs one control for each recurring requirement, such as one inventory, one pre-deployment assessment and one incident process, and records, for each control, which clauses of which instruments it satisfies. Audits then become a matter of producing evidence from a single system rather than reconciling several.
What each instrument asks
NIST AI Risk Management Framework 1.0. Released by the US National Institute of Standards and Technology on 26 January 2023 and "intended for voluntary use", the framework organises AI risk work into four functions: Govern, Map, Measure and Manage. NIST added a Generative AI Profile (NIST-AI-600-1) on 26 July 2024, and its website states that AI RMF 1.0 is being revised under the White House AI Action Plan. NIST has no legal force in India, but it is the reference vocabulary for many multinational customers and for much of the global vendor ecosystem.
ISO/IEC 42001:2023. Published by ISO on 18 December 2023, the standard "specifies requirements for establishing, implementing, maintaining, and continually improving" an AI management system. ISO describes it as designed "for entities providing or utilizing AI-based products or services", across industries and sizes, and as a management system standard implemented using the Plan-Do-Check-Act method. Where NIST describes what good AI risk practice looks like, ISO/IEC 42001 describes the management system that makes the practice repeatable and auditable.
Digital Personal Data Protection Act 2023 and Rules 2025. The Rules were notified by MeitY on 14 November 2025. According to SCC Online's summary, a small set of provisions took effect immediately, the consent manager rule takes effect on a later notified date, and most substantive rules apply eighteen months after publication. Significant data fiduciaries face additional obligations under Rule 13, including annual data protection impact assessments, independent audits and, in SCC Online's summary, a duty to verify that algorithms and software used for processing do not cause harm; PIB's explainer describes this as following "stricter checks while using new or sensitive technologies". Specified personal data and related logs may not leave India except as permitted. PIB notes that the highest penalty, up to ₹250 crore, applies to a failure to maintain reasonable security safeguards. The DPDP framework is binding law and applies to AI systems whenever they process digital personal data.
RBI FREE-AI Committee report. The Reserve Bank of India released the report of its committee on a Framework for Responsible and Ethical Enablement of Artificial Intelligence on 13 August 2025. The committee was chaired by Dr Pushpak Bhattacharyya of IIT Bombay. It sets out seven sutras, namely Trust is the Foundation, People First, Innovation over Restraint, Fairness and Equity, Accountability, Understandable by Design, and Safety, Resilience and Sustainability, and 26 recommendations under six pillars: Infrastructure, Policy and Capacity for innovation enablement, and Governance, Protection and Assurance for risk mitigation. Recommendations include a board- approved AI policy for regulated entities, governance across the AI lifecycle, extension of product approval, consumer protection and audit processes to AI, strengthened incident reporting with an indicative AI incident reporting form, disclosure to consumers when they are dealing with AI, a more tolerant compliance approach for low-risk AI, and an AI Innovation Sandbox. Outlook Money's report on the release adds an inventory of deployed AI systems, periodic audits and AI disclosures in annual reports under the Assurance pillar. These are committee recommendations; regulated entities should check the current status of any RBI directions issued since.
RBI draft guidance on model risk management. In June 2026 the RBI released a draft "Guidance on Regulatory Principles for Model Risk Management, 2026", reported by ThePrint on 26 June 2026, with comments due by 24 July. As reported, it would apply to 11 categories of regulated entities, including banks and NBFCs, and to any tool whose output feeds a business decision, "irrespective of whether such tools are recognised as models". It proposes a board-approved model risk management framework, a central model register without which a model cannot be used, with decommissioned models retained on it for at least 10 years, risk tiers reviewed at least yearly, independent validation within a three-lines-of-defence structure, testing for hallucinations and adversarial conditions, "kill- switch arrangements" to override or deactivate models, and accountability for vendor-supplied models backed by audit, access and exit rights in contracts. It is a draft; entities should check whether final guidance has been issued.
SEBI consultation on AI and ML. SEBI issued a consultation paper on guidelines for the responsible use of AI and ML in the securities market dated 20 June 2025. Taxmann's summary describes principles of equality, accountability, transparency, and safety and reliability, disclosure of AI use in areas such as algorithmic trading and advisory services, and detailed records of model development, testing, deployment and monitoring. It is a consultation, not final rules.
The unified control map for an AI risk management framework
The table below is this article's analytical contribution. It defines twelve control domains that together form one AI risk management framework, and shows for each domain where it draws on, or helps satisfy, the five instruments. A cell marked "not addressed" means the instrument does not speak to the control in the material reviewed, not that the control is unnecessary. NIST references are to functions and, where opened, to specific subcategories; ISO references are to elements of the management system rather than clause numbers, which should be confirmed against the licensed text.
Three features of the map deserve comment.
First, the inventory is the foundation. Nearly every other control depends on knowing which systems exist. Enterprises that start with policy documents and leave the inventory until later find that their policies govern a portfolio they cannot describe.
Second, the DPDP column is the only one with general binding force today, and it bites through data rather than through AI as such. A model that processes no personal data sits largely outside it; a model that drafts customer letters sits squarely within it. That makes control domain 5 the point where the AI risk framework and the data protection programme must be the same programme, not two.
Third, the sector instruments add emphasis rather than new categories. RBI's committee and SEBI's consultation both stress disclosure and records, and the RBI's 2026 draft on model risk management reinforces domains 2, 3, 6 and 10 with a mandatory model register, annual tier reviews, vendor accountability and kill-switch arrangements. An enterprise that has built domains 2, 8 and 9 well will find that most sector requests are requests for evidence it already holds.
Risk tiering: not every AI system needs every control
Applying all twelve domains at full strength to a meeting-notes summariser and to an automated credit decision is wasteful and, by consuming review capacity, makes the high-risk systems less safe. The RBI committee's recommendation of a more tolerant approach for low-risk AI points in the same direction. The tiering below is an original classification, offered as a starting point to be adapted.
Two rules keep the tiering honest. A system's tier is set by its most consequential use, not its typical use. And the tier is reviewed whenever the system's data, autonomy or audience changes, because the most common route from Tier 1 to Tier 3 is gradual scope creep rather than a single decision.
An implementation sequence keyed to the DPDP timetable
With the DPDP Rules notified on 14 November 2025 and most obligations applying after eighteen months, enterprises have a natural deadline in mid-2027. The sequence below uses that window.
First quarter: inventory and policy. Build the AI inventory across business units, including AI features embedded in purchased software. Draft and approve the AI policy, naming the accountable executive and the risk appetite. Assign each system a provisional tier.
Second quarter: assessment and data. Run pre-deployment, or for existing systems retrospective, assessments for all Tier 2 and Tier 3 systems. Align the assessment template with the data protection impact assessment so that one document serves both. Review vendor contracts for model providers against domain 6.
Third and fourth quarters: operation. Put production monitoring and the incident process in place for Tier 2 and Tier 3 systems. Separate PanScience Research pieces cover AI observability and evaluation before deployment. Define human override points and user disclosures.
Fifth and sixth quarters: assurance. Bring AI systems into the internal audit plan. Run a management review. For enterprises that expect to be designated significant data fiduciaries, or that are pursuing ISO/IEC 42001 conformity, commission an independent review of Tier 3 systems before the DPDP obligations apply.
Who owns what
A framework without owners is a document. Responsibility usually divides as follows. The board approves the AI policy and receives quarterly reporting on Tier 3 systems and material incidents. A named executive, often the chief risk officer or chief technology officer, owns the framework. The data protection officer owns domain 5 and the link to DPDP obligations. The chief information security officer owns the security aspects of domains 9 and 10. Each AI system has a business owner accountable for its outcomes and a technical owner accountable for its operation. Internal audit owns domain 11. The operating model for these roles is discussed in a separate PanScience Research piece on AI operating models.
Where frameworks usually fail
Four failure patterns recur. The inventory is built once and never updated, so new systems, especially AI features switched on inside existing software, escape review. Assessments are completed as forms before launch and never revisited, so a system's tier drifts upward unnoticed. Incident processes exist on paper but no one is told what counts as an AI incident, so near-misses go unrecorded. And the data protection and AI risk programmes are run by different teams with different templates, so the same system is assessed twice and the gap between the assessments is where the risk sits.
Conclusion
Indian enterprises do not need to choose between NIST, ISO/IEC 42001, the DPDP framework and sector guidance. They need one AI risk management framework whose controls are designed once and mapped to each instrument. The twelve-domain map in this article, combined with a three-tier risk classification, gives a practical structure: build the inventory first, merge AI assessment with data protection assessment, scale controls to risk, and use the DPDP timetable to sequence the work. The instruments will continue to evolve, with NIST revising its framework and Indian regulators developing their own positions. An enterprise with a single, mapped framework can absorb those changes by updating a column rather than starting a new project.
Methodology and sources
This article is based on primary and secondary sources opened by the author. NIST AI RMF details are from nist.gov (nist.gov/itl/ai-risk-management-framework) and the AI RMF Core (airc.nist.gov/airmf- resources/airmf/5-sec-core). ISO/IEC 42001 details are from iso.org (iso.org/standard/42001); clause- level mapping should be confirmed against the licensed standard. DPDP Rules details are from PIB's explainer (static.pib.gov.in/WriteReadData/specificdocs/documents/2025/nov/doc20251117695301.pdf) and SCC Online's summary (scconline.com/blog/post/2025/12/26/digital-personal-data-protection-rules-2025- key-highlights). RBI FREE-AI details are from the RBI report page (rbi.org.in/Scripts/PublicationReportDetails.aspx?ID=1306), the RBI press release of 13 August 2025 (rbi.org.in/Scripts/BS_PressReleaseDisplay.aspx?prid=61018) and Outlook Money's report (outlookmoney.com/banking/rbi-panel-proposes-ethical-ai-framework-for-financial-sector). The RBI draft model risk management guidance is described from ThePrint's report of 26 June 2026 (theprint.in/tech/a-human-firmly-in-the-loop-rbi-proposes-kill-switch-for-ai-used-by-banks- nbfcs/2970201); the author did not open the draft itself. SEBI consultation details are from Taxmann's summary (taxmann.com/post/blog/sebi-proposes-ai-ml-governance-framework-for-securities-markets). The unified control map, the three-tier classification, the implementation sequence and the failure patterns are the author's original analysis. Cells marked "not addressed" reflect the material reviewed and are not legal conclusions. This article is not legal advice, and the status of RBI and SEBI positions should be checked at the time of reading.