Bangladesh National Artificial Intelligence Policy 2026-2030 (draft)

What it is

The National Artificial Intelligence Policy 2026-2030 is the draft national AI policy for Bangladesh. The ICT Division published Draft V1.1 on 25 January 2026 and ran an online public consultation from 26 January to 8 February 2026. Draft V2.0 followed on 9 February 2026.

The draft has not been adopted. The ICT Division’s AI policy site lists Draft V2.0 as the latest version and says the committee is reviewing submissions before the final policy.

Who it applies to

As drafted, the policy would apply to AI across government and the private sector, including foreign AI products used in Bangladesh. Some parts are directions to government, such as procurement, the national Bangla language model and public awareness. Others describe obligations the government intends to place on developers and deployers, mainly through the risk tiers, impact assessments and liability rules. Those obligations would need law or regulation behind them to bind anyone.

What it requires

The proposals that matter most for organisations using AI are these.

• Risk tiers. AI systems would be classed as prohibited, high-risk, limited-risk or low-risk. High-risk examples include law enforcement, biometric identification, creditworthiness, welfare allocation, employment and access to essential public services. They would carry data quality, risk management, human oversight and transparency duties.
• Prohibited uses. Social scoring, indiscriminate biometric mass surveillance in public spaces without lawful authorisation, manipulative or deceptive AI that exploits vulnerabilities, and AI used to interfere in elections.
• Algorithmic impact assessments. Mandatory before deployment for significant public sector AI and for high-risk AI deployed by private actors, with documentation available to regulators and, where appropriate, the public.
• Rights over automated decisions. A right to a counterfactual explanation of the key factors behind a high-stakes automated decision, and a right to review by a qualified person able to change or override the outcome.
• Children. Consent from a parent or guardian before AI companies collect data of children under 16, and bans on profiling, behavioural tracking and AI-targeted advertising aimed at children, and on using children’s data to train AI models.
• Liability and evidence. Strict liability for harm caused by high-risk AI in sectors such as health, finance, employment, justice, welfare, education and critical infrastructure. Deployers of high-risk AI would keep model versions, training and inference logs, decision records and system parameters for at least five years.
• Institutions. The national data governance authority would coordinate AI governance, issue technical standards, certify high-risk systems and oversee impact assessments. An independent oversight committee would monitor AI deployments and publish an annual report. Regulatory sandboxes would allow time-limited testing.

Key dates

• 25 January 2026. Draft V1.1 published.
• 26 January to 8 February 2026. Online public consultation.
• 9 February 2026. Draft V2.0 published.

The draft proposes a mid-term review in 2028 and a sunset on 31 December 2030 unless the policy is renewed or replaced by an AI law.

How it relates to other frameworks

The risk tiers and prohibited practices follow the shape of the EU AI Act. The draft names the data protection law as the base for accountability, and the Personal Data Protection Act 2026 already lets people withdraw consent to automated decisions. It relies on the Cyber Security Act 2026 to hold people accountable for harmful AI use.

An organisation that keeps an AI inventory with risk levels, runs impact assessments before deployment, records human oversight for consequential decisions and keeps logs and decision records will already hold most of what the draft would ask for. Mapping that work to the NIST AI RMF or ISO/IEC 42001 keeps it reusable whatever the final text says.

A note on this page

This is Aicura’s reading of Draft V2.0 of the policy. It is not legal advice, and a draft can change before adoption. Read the current text from the ICT Division and take advice on how Bangladeshi AI policy applies to your organisation.