Dr. Aditya Vikram Kashyap · AI governance researcher and practitioner
I study how institutions govern artificial intelligence as it moves from producing information to exercising authority. Regulated financial systems are my empirical laboratory. The question underneath is broader: how any institution keeps accountable control as AI systems acquire greater autonomy, delegated authority, and infrastructural importance.
Dr Aditya Vikram Kashyap is a financial services executive and AI governance researcher specializing in institutional AI governance, agentic AI, and the governance of increasingly autonomous systems. He brings more than a decade of experience in regulated finance to questions that increasingly command the attention of boards and executive teams: how artificial intelligence changes competitive advantage, institutional capability, and accountability.
As a Vice President of Firmwide Innovation at Morgan Stanley, his work has spanned the evaluation of emerging companies and technologies, build-versus-buy and licensing decisions, product incubation, and the governance of new capabilities. His experience sits at the intersection of the decisions that determine what organizations build, acquire, deploy, and govern.
Aditya is a Senior Fellow at the Institute for Ethics in Artificial Intelligence (IEAI) at the Technical University of Munich and a Visiting Senior Fellow with the Media, Technology & Society Programme at ISEAS – Yusof Ishak Institute. His research examines the consequences of increasingly autonomous AI for markets and institutions, with particular attention to governance, institutional resilience, technological dependence, and accountability. His doctoral research produced the KARMA Framework for institutional AI governance, grounded in the realities of regulated financial services. He is also a Working Group Member of the IEEE P7999 Series within the IEEE Standards Association, contributing to standards work on AI ethics oversight.
He has published more than 50 articles and commentaries on artificial intelligence, AI governance, financial services, and emerging technology, including work for Forbes, CNN-News18, CNBC-TV18, and India Today. Across his research and writing, he connects the strategic choices facing organizations with broader questions of power, responsibility, and trust raised by increasingly capable AI systems.
Aditya holds an Executive Doctorate in Business Administration from Saint Mary's University, a master's degree from New York University, and a bachelor's degree from Drexel University. He is a Fellow of the Institution of Engineering and Technology (FIET), BCS, The Chartered Institute for IT (FBCS), and the Institution of Electronics and Telecommunication Engineers (FIETE), as well as an IEEE Senior Member.
His research and writing are conducted in an independent capacity. The views expressed here are his own and do not represent those of any affiliated institution.
The problem
Many governance mechanisms still assume a system whose primary role is to produce information or recommendations. A model produced an estimate; a person or an accountable process retained the decision.
That arrangement is dissolving. Systems now initiate actions, coordinate workflows, call other systems, and exercise delegated authority under conditions that may not have been examined at the point of approval. When that happens the governing question is no longer only whether the model is accurate. It becomes who authorized the action, on what evidence, under what constraints, and whether that authorization still holds.
AI can reshape authority before it improves performance. The moment a system is permitted to decide, power has moved inside the organization, whether or not the move was deliberate.
Governance mechanisms built around point-in-time approval and scheduled validation become vulnerable when systems can change materially after deployment. Fragmented accountability can allow consequential behavior to escape clear ownership; excessive centralization can make governance itself the bottleneck. The institutional problem is therefore not simply how much governance exists, but whether authority, evidence, monitoring, and intervention remain aligned as the system changes.
Closing that distance is the work I do.
I study what happens when software is permitted to act on its own inside an institution that has to answer for what it does. The hard problem is no longer only whether a model is accurate. It is what authority the system has been granted, who remains answerable for its actions, and what evidence justified that grant.
AI governance is not short of principles, standards, or normative frameworks. What is thinner is the layer between principle and institutional operation: decision rights, approval authority, monitoring obligations, change triggers, escalation paths, evidence standards. That layer is where my research sits.
I work on it as a practitioner as well as a researcher, which is a test rather than a credential. A governance idea that cannot tell a bank manager what to do on a Monday morning has not finished being an idea.
The same agenda now runs through several institutions in Singapore, Germany, Brazil, and India. They are listed below. Their value is that the questions get tested against other people's evidence rather than only my own.
Current appointments
The same questions are pursued through a small number of institutions. Research appointments are listed first, followed by standards and governance service.
Research propositions
Six claims that organize the research. Each is contestable, each is argued at length somewhere in the writing, and each would be worth abandoning if the evidence went the other way.
AI governance is becoming the governance of delegated authority rather than the governance of models.
The Architecture Of PermissionApproval cannot be a fixed point when the system it approved is able to change after approval.
The Most Dangerous AI Risk Isn't Hallucination, It's Behavioral DriftExplainability and accountability are different properties. Treating one as a substitute for the other is a category error.
The Dangerous Illusion Of Explainable AI In Modern FinanceThe unit of governance is moving from the model to the architecture assembled around it.
AI's next phase isn't a model. It's an architectureDepending on a capability an institution does not own is a governance problem, not merely a procurement one.
You Don't Control The Infrastructure Your Bank Runs OnEfficiency and resilience optimize for different states of the world. Retaining the ability to choose later can be worth more than optimizing now.
The Strategic Value Of OptionalityResearch program
By institutional architecture of AI power I mean the arrangement of authority, evidence, accountability, and control through which an organization decides what a machine may do, on whose behalf, under what limits, and who can stop it. It is the layer beneath the model and above the policy document, and it is where I think the governing problems now sit.
Four questions organize the research. They are not service categories. They remain unsettled across research, policy, and institutional practice, and they recur wherever an institution must answer for what a machine has done.
Regulated financial systems provide my primary empirical setting, because that is where delegated authority, supervisory scrutiny, and technological dependence already meet. The research question is not confined to banking. My doctoral research examined the governance approaches published by major AI developers, among them IBM, Microsoft, Google DeepMind, OpenAI, and Anthropic, alongside the institutional controls used in regulated financial systems. The comparison directed my attention to a recurring institutional question: how authority, evidence, accountability, and the power to intervene are assigned as AI systems become more consequential.
Delegated machine authority is the power an institution grants a system to act on its behalf. What decisions, actions, or institutional powers may be delegated, who authorizes that delegation, and what constraints attach to it?
The Architecture Of Permission CNN News18, July 2026When has a system, its behavior, its architecture, or the conditions of its use changed enough that an earlier authorization should expire, be reconsidered, or require renewed evidence?
The Most Dangerous AI Risk Isn't Hallucination, It's Behavioral Drift Forbes, April 2026How should responsibility be assigned when consequential outcomes emerge across people, models, agents, vendors, data, infrastructure, and automated workflows rather than from a single identifiable decision-maker?
The Dangerous Illusion Of Explainable AI In Modern Finance Forbes, January 2026How does an institution retain meaningful control over capabilities it relies on but does not own, including the infrastructure, evidence, operating conditions, and intervention rights on which that control depends?
You Don't Control The Infrastructure Your Bank Runs On Forbes, June 2026The research engages with governance instruments that shape institutional practice, including the NIST AI Risk Management Framework, ISO/IEC 42001, the EU AI Act, and Federal Reserve model risk guidance.
Research contribution
KARMA is a governance architecture developed through my doctoral research to translate AI governance principles into institutional operating mechanisms. It organizes the problem into four questions an institution should be able to answer before it permits a system to act, and maps each onto the machinery that answers it: what is known about the system, who holds the decision right, who remains answerable, and what constrains the system itself.
What can this system actually do?
What it can do, what it cannot, what could go wrong, and what evidence supports each of those answers. Tracked from build to retirement rather than established once.
Who may authorize it, and who may stop it?
What authority is being delegated, what limits attach to it, who can constrain or terminate it, and when must it be renewed rather than assumed to persist.
Who answers for the outcome?
Responsibility is the duty to act with care. Accountability is the obligation to answer afterwards. A system can have designated owners while still leaving consequential outcomes without a sufficiently clear answerable party. That is the failure the architecture is designed to prevent.
What constrains the system itself?
What is tested before release, what is monitored afterwards, what triggers escalation without a person having to notice, and what the system is technically prevented from doing.
The KARMA Framework is one answer emerging from the wider research program, developed for regulated financial institutions while addressing institutional questions that recur beyond the sector. Originated by Aditya Vikram Kashyap. The framework is associated with Indian Patent Application No. 202611072022. The application is pending; no grant is claimed.
Research perspective
He brought to the work a rare combination of rich technology management domain expertise, deep financial services knowledge, and a sophisticated understanding of AI governance.
His research moves beyond high-level principles and develops a practical, deployer-oriented governance framework for financial institutions.
Selected essays
Sequenced by argument rather than by date. They move from why the governable unit is shifting, through authority and behavior, to accountability, dependence, and what resilience requires. Each line states the claim the piece makes.
Forbes·India Today·CNN News18·CNBC-TV18·ISEAS
Everything published since 2025 on AI governance, model integrity, institutional resilience, and technology policy, including research, policy analysis, commentary, and opinion writing.
Selected talks and conversations
Open Source in Finance Podcast, FINOS, 2026
A conversation on innovation under regulation and the institutional conditions that allow technological change to proceed responsibly in financial services.
These are the primary records I use for identity and citation verification.
Engagement
The work is most useful where questions of AI capability become questions of institutional authority, accountability, and control. I collaborate selectively with research institutes, policy organizations, standards bodies, boards, academic centers, and regulated institutions on those questions, through research, governance review, closed-door discussion, and public engagement.
Joint research, think tank programs, institutional studies, and framework development on AI governance, delegated authority, and accountability.
Discuss a research questionBoard and senior-leadership briefings, governance reviews, and structured discussion of AI oversight, delegated authority, model change, and institutional control.
Discuss a governance questionKeynotes, lectures, panels, roundtables, and expert commentary on the institutional consequences of increasingly autonomous AI systems.
Invite me to speakGoes straight to my inbox.
For conference programs, contributor notes, panel introductions, and media use, the biographies below may be used as written.
One line
AI governance researcher and practitioner studying the institutional architecture of AI power.
Short, about 80 words
Dr. Aditya Vikram Kashyap is an AI governance researcher and practitioner. His work examines how institutions retain accountable control when artificial intelligence moves from producing information toward exercising consequential authority. He is a Visiting Senior Fellow with the Media, Technology & Society Programme at ISEAS – Yusof Ishak Institute, a Senior Fellow with the Institute for Ethics in Artificial Intelligence at the Technical University of Munich, and a Visiting Scholar at FGV Direito Rio. He originated the KARMA Framework for institutional AI governance.
Standard, about 125 words
Dr. Aditya Vikram Kashyap is an AI governance researcher and practitioner. His work examines how institutions retain accountable control when artificial intelligence moves from producing information toward exercising consequential authority, with recurring attention to delegated machine authority, agentic AI, institutional accountability, system change, technological dependence, and resilience. He is a Visiting Senior Fellow with the Media, Technology & Society Programme at ISEAS – Yusof Ishak Institute, a Senior Fellow with the Institute for Ethics in Artificial Intelligence (IEAI) at the Technical University of Munich, and a Visiting Scholar with the Center for Technology and Society and CyberBRICS at FGV Direito Rio. Regulated financial systems provide his primary empirical setting. He originated the KARMA Framework for institutional AI governance, and writes for Forbes, India Today, CNN News18, and CNBC-TV18.
Long, about 255 words
Dr. Aditya Vikram Kashyap is an AI governance researcher and practitioner. His work examines how institutions retain accountable control when artificial intelligence moves from producing information toward exercising consequential authority, with recurring attention to delegated machine authority, agentic AI, institutional accountability, system change, technological dependence, and resilience. He is a Visiting Senior Fellow with the Media, Technology & Society Programme at ISEAS – Yusof Ishak Institute, a Senior Fellow with the Institute for Ethics in Artificial Intelligence (IEAI) at the Technical University of Munich, and a Visiting Scholar with the Center for Technology and Society and CyberBRICS at FGV Direito Rio. Regulated financial systems provide his primary empirical setting. He originated the KARMA Framework for institutional AI governance, and writes for Forbes, India Today, CNN News18, and CNBC-TV18. His research at the Center for Technology and Society concerns AI governance, agentic systems, financial services, and institutional accountability; his work with CyberBRICS contributes to research and policy dialogue on AI governance and digital transformation across the BRICS countries and the broader Global South. He holds an Executive Doctorate of Business Administration from Saint Mary's University, where his research examined the governance approaches published by major AI developers alongside the institutional controls used in regulated financial systems. He is a Fellow of the Institution of Engineering and Technology, BCS The Chartered Institute for IT, and the Institution of Electronics and Telecommunication Engineers, and a Senior Member of the IEEE. He is also a Working Group Member of the IEEE P7999 Series within the IEEE Standards Association and a FINOS Ambassador.
Media and conference, about 75 words
Dr. Aditya Vikram Kashyap researches the institutional architecture of AI power: how organizations keep accountable control as artificial intelligence begins to exercise authority rather than only produce information. He is a Visiting Senior Fellow at ISEAS – Yusof Ishak Institute, a Senior Fellow at the Institute for Ethics in Artificial Intelligence at the Technical University of Munich, and a Visiting Scholar at FGV Direito Rio. He originated the KARMA Framework for institutional AI governance.
ORCID 0009-0006-1770-3601. The views expressed here are his own and do not represent those of any affiliated institution.