Close Bangladesh's AI Policy Vacuum: A National AI Strategy and Governance Framework Led by the ICT Division

Diagnosis

Bangladesh has no AI strategy and no governance framework. That is the core finding here, and it matters now because AI deployment does not wait for policy. Procurement of AI tools by ministries, banks, and private firms proceeds regardless of whether the state has rules for data use, model accountability, procurement standards, or redress when an automated system causes harm. In a vacuum, each agency improvises its own approach, vendors set the terms, and the public bears the downside of biased, opaque, or insecure systems with no body to appeal to.

A vacuum is not neutral. It is itself a policy choice that defaults control to whoever moves first, usually foreign vendors and the better-resourced ministries, while leaving citizens, small firms, and regulators without standards to point to. The longer the gap persists, the more entrenched ad hoc practices become, and the harder a coherent framework is to impose later. The lead responsibility sits with the ICT Division (ICTD), with the Bangladesh Computer Council, the Bangladesh Hi-Tech Park Authority, and the Ministry of Science and Technology as supporting bodies.

Recommended actions

1. Issue a National AI Strategy as a Cabinet-approved policy document. Owner: ICT Division. Mechanism: a formal strategy document, drafted by ICTD and cleared through Cabinet, that sets national objectives, sector priorities, and a definition of high-risk AI uses. Observable signal: a published, dated strategy with named accountable units, not a discussion paper.
2. Stand up an interim AI governance body by administrative circular. Owner: ICT Division. Mechanism: an ICTD circular establishing a cross-government AI governance unit, chaired by ICTD with the Bangladesh Computer Council as technical secretariat, before primary legislation is ready. Observable signal: the unit meets, publishes terms of reference, and begins reviewing public-sector AI deployments.
3. Mandate an AI use register and risk-tier classification for all public-sector systems. Owner: Bangladesh Computer Council, under ICTD direction. Mechanism: a registration requirement for every government AI deployment, classified by risk tier, maintained by the Council. Observable signal: a populated register that grows as ministries report systems, with high-risk uses flagged for review.
4. Embed AI procurement and assurance standards into government contracting. Owner: ICT Division with the Bangladesh Hi-Tech Park Authority. Mechanism: standard procurement clauses requiring documentation, audit access, and accountability terms in any AI contract, applied through existing public procurement channels and Hi-Tech Park tenancy conditions. Observable signal: new AI contracts carry the standard clauses and vendors accept audit obligations.
5. Build state capacity and a research base through the Ministry of Science and Technology. Owner: Ministry of Science and Technology with ICTD. Mechanism: a funded skills and applied-research programme to train civil servants who can evaluate AI systems and to seed domestic technical talent. Observable signal: trained reviewers staffing the governance unit and a pipeline of assessed systems.

Sequencing (first 12 months)

Start with the strategy document and the interim governance circular, because both can be issued under existing ICTD authority without waiting for new law. These two unlock everything else: the strategy gives the risk definitions, and the governance unit gives the body that runs the register, clears procurement standards, and commissions the capacity programme. The use register follows once the unit exists, then the procurement clauses, then the capacity build. Primary legislation is the year-two destination, not the year-one starting point.

Risks and constraints

The binding constraints are institutional, not technical. AI governance cuts across many ministries, so ICTD must secure Cabinet backing or the strategy stays a paper. Fiscal space for a capacity programme is tight, and the governance unit risks becoming a box-ticking layer rather than a real review function. There is also a sequencing risk: moving too slowly lets ad hoc deployments entrench, while moving too fast with heavy rules could stall legitimate adoption.

Bottom line

Bangladesh's AI policy vacuum is a live liability, and the fastest credible fix is for the ICT Division to issue a national strategy and stand up an interim governance body by circular within the first year. Everything else, the use register, procurement standards, and capacity, follows from those two early moves under authority ICTD already holds.