AI Dispatch | RBI Seeks AI Kill Switch for Banks

The Reserve Bank of India (RBI) has put banks and financial firms on notice over risky AI models, proposing rules that would require them to keep human oversight and rapid shutdown controls in place. The move came in a week when Indian IT firms framed agentic AI as a $300 billion to $400 billion services opportunity, OpenAI appointed its India head, and India-US talks turned again to chips, compute and critical minerals. Outside India, the warnings hardened, with the UN and BIS pointing to weak governance and the danger of excessive AI infrastructure spending.

RBI Warns Banks on Risky AI

The Reserve Bank of India has proposed a draft Model Risk Management Framework that would bring AI and other decision-making models used by banks and financial firms under tighter supervision. The draft says regulated entities should be able to override, pause or deactivate a model quickly if it produces faulty, harmful or unreliable results. It also calls for human oversight, independent validation, risk classification and board-level accountability.

The framework would apply to models built in-house as well as those sourced from vendors. For banks, the point is clear: once AI starts shaping credit, risk, customer or operational decisions, it cannot be left to run as a black box.

Indian IT Firms See $400 Billion AI Opening

India’s technology services firms are trying to move AI work from pilots into live operations, as clients seek help with data, governance and workflow redesign. According to Nasscom, about 85% of providers now use agentic AI platforms, while nearly a quarter have moved AI projects from pilots to production. At the Nasscom US Forum at the Consulate General of India in New York, industry leaders said agentic AI could open $300 billion to $400 billion in new tech services spending by 2030.

The pitch is shifting from adding AI tools to reworking how jobs are assigned, checked and completed. That is where Indian IT firms see the bigger services opportunity.

OpenAI Names India Head

​OpenAI has appointed Prabhjeet Singh, Uber’s former president for India and South Asia, as Managing Director for India. Singh will start in September and report to Kiran Mani, OpenAI’s Managing Director for Asia Pacific. He will lead OpenAI’s India operations, including user growth, enterprise adoption, partnerships, policy engagement and regulatory outreach.

The appointment gives OpenAI a senior local executive in one of its fastest-growing markets. India is becoming more important for AI companies as a consumer market, developer base and enterprise sales opportunity, even as questions around pricing, data use, local infrastructure and regulation remain unsettled.

​India-US Talks Turn to AI Stack

India and the US held a closed-door industry roundtable in Washington last week on artificial intelligence, semiconductor supply chains and critical minerals. The meeting was organized by the Indian embassy in Washington with the US-India Strategic Partnership Forum and Silverado Policy Accelerator. It brought together officials, policymakers and firms working on AI, chips, quantum technologies and critical minerals.

The discussion showed how AI cooperation is moving beyond models and applications to the hardware and supply chains underneath them: compute, semiconductors, minerals, energy and secure manufacturing capacity.

Global

UN Panel Warns on AI Risks

A United Nations-backed scientific panel has warned that artificial intelligence is advancing faster than governments’ ability to test, measure or regulate it. The warning appears in the preliminary report of the Independent International Scientific Panel on Artificial Intelligence, a 40-member body created by the UN General Assembly in 2025 to assess AI’s opportunities, risks and wider impact.

The report flags risks across cybersecurity, biotechnology, disinformation, financial systems and democratic institutions. It says governments still lack the evidence, testing capacity and oversight tools needed for advanced AI systems, particularly as more agentic systems move into real-world use.

​Sovereign Funds Chase AI Infrastructure Boom

Sovereign wealth funds are putting more money into infrastructure, private credit and private equity as the AI boom changes how large state investors seek returns. Invesco’s 2026 Global Sovereign Asset Management Study surveyed 144 senior investment professionals from 90 sovereign wealth funds and 54 central banks managing about $29 trillion in assets.

The report found rising interest in energy security and transition infrastructure, helped by demand from AI data centers and other compute-heavy assets. Infrastructure accounted for 9% of sovereign wealth fund assets in 2026. The shift also shows a move away from crowded public equity markets and toward assets tied to long-term demand for power, data and digital infrastructure.

​OpenAI, Anthropic Target Science

Anthropic and OpenAI have announced separate research tools aimed at computational biology and life sciences. Anthropic launched Claude Science, a beta workbench designed to help researchers review literature, analyze data, run code and manage scientific computing workflows. OpenAI introduced a benchmark to test whether AI models can handle computational biology tasks that require scientific judgment rather than fixed workflows.

The releases show AI labs pushing beyond general-purpose chatbots into more specialized scientific work. The value could be significant, but only if the tools prove reliable in settings where errors are harder to forgive.

BIS Warns on AI Capex Bust

The global race to build AI infrastructure could become a prolonged investment bust if companies spend too far ahead of demand, the Bank for International Settlements warned in its Annual Economic Report 2026. The BIS said AI could raise productivity over the coming decade, noting that task-level studies often show time savings of 20% to 50%. But it warned that heavy spending, high expectations and concentrated investment by large technology companies could create financial risks if returns fall short.

The risk is not that AI fails outright. It is that infrastructure spending may run ahead of demand before productivity gains become visible. For all the excitement around compute, investors still need proof that the capacity will be used profitably.