SBP Finalizes AI Guidelines for Financial Sector

The State Bank of Pakistan (SBP) is finalizing critical guidelines to ensure the responsible use of artificial intelligence (AI) in the financial services sector, prioritizing trust, transparency, and accountability, while staunchly protecting consumer rights and interests.

The Financial Stability Review 2024 underscores the rapid adoption of AI technologies in Pakistan’s financial institutions, mirroring global trends. Local banks are leveraging robotics, process automation, virtual assistants, and machine learning for tasks like fraud detection, risk management, customer service, and operational efficiency. A survey of 55 regulated entities (REs)—including commercial banks, microfinance institutions, digital banks, and payment service operators—revealed that nearly half have either deployed AI or are actively developing it for applications like credit risk assessment, marketing, and automation.

However, the SBP stresses the urgent need to address environmental risks tied to AI. Banks must integrate the carbon footprint of AI systems into their risk management frameworks, treating it as a distinct risk category. AI in banking, especially for real-time processing and high-accuracy tasks, is energy-intensive, necessitating emission measurements across the AI lifecycle and the adoption of energy-efficient models to curb environmental impact.

The report aligns with International Financial Reporting Standards (IFRS) S1 and S2, which mandate sustainability disclosures, including climate-related risks, being rolled out in Pakistan for listed companies based on assets, turnover, and employee numbers. The SBP urges banks to define clear risk appetites and tolerance levels for AI’s carbon footprint, balancing operational efficiency with environmental responsibility.

Additionally, the SBP highlights systemic risks from AI, such as operational failures, cyber-attacks, and supply chain disruptions, alongside the risk of market correlations from widespread use of similar AI models, which could amplify asset price vulnerabilities.