Don't use confidential data in AI tools: Bangladesh Bank

The Bangladesh Bank has instructed its officials and employees not to input any confidential or sensitive banking information into artificial intelligence (AI) tools under any circumstances, citing growing concerns over data security and cross-border information transfer.

The directive, issued through an office order on 28 June, also requires prior approval from the appropriate authority before using AI tools for official tasks such as drafting office notes, formulating policies or analysing data.

According to the order, the use of AI platforms such as ChatGPT, Gemini, Claude, Grok and DeepSeek has recently increased among its officials for tasks including office noting, data analysis and report preparation.

Using such tools could result in sensitive and confidential financial information being transferred outside the country, creating risks of data leakage, policy breaches and serious security concerns.

To address these risks, Bangladesh Bank issued four key instructions: officials must not input confidential or sensitive office information into AI tools; they must obtain prior approval before using AI for official work such as note drafting, policy formulation or data analysis; they must not use office-related information even when using AI tools for personal purposes; and they must strictly comply with existing information security and cyber risk management policies.

The central bank said the directive aims to ensure the responsible use of AI technology while safeguarding information security, and that it has been issued with the approval of the competent authority.