Kailash Sadangi warns AI finance controls are lagging adoption

Finance governance researcher Kailash Sadangi says AI is spreading faster through corporate finance than the controls needed to manage model risk. His analysis points to a widening gap between AI use, governance maturity and regulatory oversight as finance teams deploy more autonomous tools.

Why it matters: - AI is now embedded in core finance work, but model risk controls are not keeping up. - The gap raises the chance that finance teams will miss drift, hallucinations or other model failures in planning, reporting and analysis. - Governance failures could leave firms relying on outputs that are not covered by the same checks used for human-generated data.

What happened: - Finance governance researcher Kailash Sadangi said AI adoption in finance is outpacing internal control frameworks. - His analysis draws on KPMG's 2026 global survey, which found active AI use across finance has more than doubled in two years. - More than three-quarters of organisations now use AI in financial planning, reporting and commercial analysis. - KPMG also found 71% of organisations said AI is meeting or exceeding return-on-investment expectations. - Sadangi's research says deployment has moved faster than the governance systems built to oversee it.

The details: - BCG research on financial institutions found 71% rated their AI capabilities at mid-tier maturity or above. - Objective assessment showed only around 25% had truly integrated AI into strategic operations. - Sadangi identifies that mismatch as a governance problem, not just a technology gap. - Internal controls, audit trails and sign-off processes built for human-generated financial data do not automatically extend to model-generated outputs. - The risk grows as generative AI and agentic AI shift from single-output tools to more autonomous, multi-step decision chains. - In April 2026, the US Federal Reserve, the OCC and the FDIC issued SR 26-2, a major update to model risk management guidance that had stayed largely unchanged for more than a decade. - The updated guidance explicitly excludes generative and agentic AI from its formal scope because the technologies are changing so quickly. - Separate global research on generative AI in financial institutions says firms need to document AI use cases, conduct model audits and add human oversight without waiting for unified global regulation. - Research on AI-driven cyber threat intelligence in finance found recurring problems including shadow use of AI tools outside formal controls, weak security monitoring and missing audit-ready evidence for AI models. - Sadangi said the least-audited part of the process is often the one finance teams rely on most. - "The risk isn't that AI in finance produces obviously wrong numbers — most of the time it doesn't," Sadangi said. - "The risk is that when a model drifts, hallucinates or degrades in accuracy, the controls designed to catch human error may not be built to catch model error, and few finance functions have clearly assigned who is accountable for closing that gap," Sadangi said. - He said closing the gap does not require finance leaders to become AI engineers. - He said firms need model audits treated as seriously as financial audits. - He said AI-assisted workflows should include documented human sign-off points. - He said model-risk ownership should sit in one clearly assigned place rather than be spread across IT, risk and finance. - Sadangi is a senior finance executive with Group CFO experience across Australia, the Middle East and international markets. - Sadangi is also a doctoral researcher examining CFO-centred governance of AI-enabled decision-making. - Source links cited in the release include KPMG's 2026 AI in Finance survey, BCG-related AI maturity data, the Federal Reserve/OCC/FDIC model risk guidance, a global survey on generative AI in financial institutions, and research on AI-driven cyber threat intelligence in finance.

Between the lines: - Regulators appear to be acknowledging that current model-risk frameworks were built for a slower technology cycle. - The exclusion of generative and agentic AI from SR 26-2 suggests policy is still chasing day-to-day deployment. - The core issue is not whether AI can help finance teams. It is whether governance can verify, explain and sign off on what the models do.

What's next: - Sadangi's analysis points to more pressure on finance leaders to define model ownership, document use cases and add formal review steps. - Firms using generative or agentic AI in finance are likely to face stronger demands for audits, human oversight and evidence trails. - As AI tools become more autonomous, model-risk management will likely become a board-level governance issue rather than a niche technology control.

The bottom line: - AI adoption in finance is moving faster than the systems designed to police it, and that gap is now the risk story.