Bank Indonesia Warns AI Could Fuel Herding Behaviour and Bank Runs, Urges Sovereign Control

Bank Indonesia has told the country’s lenders to watch for a set of vulnerabilities that come with their growing use of artificial intelligence, from fraud and security gaps to models that could push the whole system in the same direction at once, Katadata reported on October 27. Ricky Satria, head of the central bank’s Data Innovation and Digitalisation Department, spoke at the launch of a white paper titled Digital Sovereign AI and the Future of Trusted Banking, published by IBM, the banking association Perbanas and K&A.

Fraud schemes are becoming more convincing as criminals use AI to clone voices and even fake faces on video calls, he said, and attacks that exploit weaknesses in banks now move in a matter of hours rather than weeks. “Attacks are very fast, no longer weeks,” Ricky said, adding that large banks overseas have found AI agents operating in their systems without full visibility, agents that could expose customer data.

The deeper concern is herding behaviour. Citing the International Monetary Fund, Ricky noted that many AI models draw on the same inputs, including financial reports, economic news and regulatory documents, so they can reach the same decisions at the same time. In calm markets that can speed up investment decisions and add liquidity, but under stress it could amplify selling, drain liquidity and erode customer confidence, and in the worst case spill over into mass withdrawals, a bank run.

What it means in practice

BI’s answer is control. Ricky said banks must stay open to innovation but keep full command of the AI they use, which is why the central bank is encouraging sovereign AI, meaning models and infrastructure that a country or institution develops, runs and governs itself. That lines up with the government’s plan to write sector-specific AI rules for finance, health and education and with the data-control push in the One Data Indonesia law covering AI and data sovereignty. For bank risk teams, it means inventorying every AI agent with access to customer data, testing models for correlated behaviour under stress, and preparing for deepfake-driven social engineering in call centres and apps.

The next step to watch is whether BI and the Financial Services Authority, OJK, turn these warnings into supervisory guidance on AI model risk and agent access for Indonesian banks.