Artificial intelligence (AI) governance must evolve from a compliance function into an operating framework that enables innovation, builds trust and allows organisations to scale responsibly, senior industry executives said during a panel discussion on responsible AI.
The discussion at BW Chief Data & AI Officer Summit 2026 featured Kritika Muthukrishnan, Chief Product Officer, Reliance Enterprise Intelligence; Jay Shah, Chief Data and Analytics Officer, Kotak Mahindra Bank; and Shubhashree Dasgupta, CDAO India, Data and Analytics Office, HSBC.
Governance As A Guardrail
Muthukrishnan said organisations are increasingly moving away from the view that governance and innovation are competing priorities. Instead, governance can provide guardrails that allow teams to innovate faster.
She said the emerging approach is to establish governance guardrails at the beginning of an AI initiative, covering explainability, data usage and other requirements, rather than treating governance as an additional process after development.
Muthukrishnan cited account aggregators in financial services as an example of governance supporting innovation. The consent-based framework gives users control over what information is shared, for how long and for what purpose. She said account aggregator-based loans had recorded about 600 per cent growth, while the overall amount of loans had doubled between FY25 and FY26.
Shah said strong data science and analytics capabilities tend to be accompanied by strong governance functions, particularly in banking. Governance, he said, should be viewed as an enabler and work closely with business and development teams.
He also stressed the importance of having practitioners within governance teams. While legal and compliance expertise is important for defining the framework, practitioners who understand AI and analytics can better engage with development teams and assess both the value and risks of use cases.
Building Trust Through Data
Dasgupta said trusted AI depends on foundational data governance covering security, quality, privacy, lifecycle management, oversight and assurance.
She highlighted the need for data to be tokenised, encrypted and access-controlled when it is shared with frontier large language models (LLMs). Such measures, she said, reduce the “surface area of attack” while allowing innovation to continue.
Dasgupta also emphasised consent, particularly when customer data is used to develop AI applications such as voice agents. Organisations must know who has access to their data and how third-party and fourth-party vendors handle it, she said.
For Shah, accountability must extend across the AI development lifecycle. He advocated moving from a late-stage approval process to “tool gates”, where governance is embedded alongside development. This allows issues to be addressed during the creation of a use case rather than after it is completed.
“Ultimately setting the principles of the governance process for AI is the most important,” Shah said. He added that organisations must understand the business purpose, underlying data, data selection, and the model development process. Auditability and maintaining records of training and execution artefacts are also critical.
Regulation And Data Sovereignty
The panel also examined the difficulty of preparing for rapidly evolving AI regulation. Muthukrishnan said regulatory frameworks across India and global markets remain divergent, with requirements emerging at different speeds.
For organisations, she said, the challenge is not only compliance but also sovereignty. Data may be stored locally, but organisations must also consider whether data or model inference leaves the country when external AI models are used.
She said organisations can build model-agnostic governance frameworks around explainability and bias while remaining flexible about which models they use. Open-weight models and smaller language models may offer greater control, although frontier models remain important for some use cases.
Dasgupta said common principles are emerging across jurisdictions around responsible AI, consent, privacy and data governance. At the same time, countries are increasingly looking to develop local AI capabilities and interoperable systems.
Leave a Reply