Israeli Financial Sector Integrates AI Agents with Customer Data Amid Regulatory Challenges
Over the past six weeks, the Israeli financial market has shifted as Bank One Zero and several financial firms, including IBI SMART and Fair, have established direct connections between customer financial data and external language models like ChatGPT and Claude. This development marks a significant step from AI serving merely as a question-answering assistant to becoming an active agent that can view, analyze, and potentially manage customers' money. Currently, these connections are limited to data viewing and insights, but discussions are underway about advancing to "Supervised Autonomy," where AI agents could initiate transfer orders subject to customer approval.
The Bank of Israel acknowledges this trend as part of a global banking shift, highlighting "Agentic AI" as a key emerging technology alongside risk management, fraud detection, and credit advisory. While still in early stages in Israel, regulators emphasize the need to establish proper AI governance frameworks in anticipation of broader adoption.
A recent incident in the global crypto sector underscores the risks involved. On August 1, the CEO of digital custody firm BitGo publicly deposited 100 bitcoins, worth approximately $6.3 million, into a public wallet and challenged Anthropic's Claude models to transfer the funds. This followed a July 30 safety report revealing that Claude models had mistakenly accessed real production systems and sensitive data due to configuration errors. BitGo's CEO stressed these were test environment issues rather than advanced breaches and noted the institutional multisignature wallet setup requires multiple keys for transfers, adding security layers.
This situation echoes Israel's prior experience with cloud computing regulation, where initial strict permissions for cloud use in banks gradually eased after practical supervised implementation. A similar pattern may unfold with AI agents: the market moves ahead with calculated risks, while regulation trails and adapts based on real-world trials. However, unlike cloud data, financial data is now exposed to foreign AI companies, and Israel currently lacks clear frameworks defining liability when AI agents err with customer financial information. The market is advancing rapidly, but questions of responsibility remain unresolved.
Michal Ohana, a partner in the financial regulation department at Shibolet Law Firm, highlights these developments and the regulatory challenges ahead.
