Bangladesh rewrote almost its entire digital statute book in April 2026, threading artificial intelligence through the definitions of its cyber and telecommunications law and criminalising the worst uses of synthetic media. None of it asks anyone who builds or deploys an AI system to do anything. There is no adopted AI policy, and the only labeling mandate in the corpus sits in a draft.
The shape is a sparse jurisdiction that leads with policy: two AI-specific instruments sit in the soft tier, no dedicated AI law exists or is tabled, and the binding tier is general law an AI dispute would meet. The distinctive weight sits in the regional band, where Bangladesh holds a confirmed founding seat in the new AI cooperation organization.
any person or software developer or user of artificial intelligence tools
Puts the user of an AI tool inside the offence, alongside the person and the developer.
committed by a person or artificial intelligence agent
An AI agent can be the committing actor, not merely the instrument.
কৃত্রিম বুদ্ধিমত্তা দ্বারা নির্মিত অথবা এডিটকৃতTwo years or 10 lakh taka; five years or 20 lakh where the victim is a wbangladesh or a child.
Searched for the deepfake, label, marking and synthetic terms. Nothing.
Analysis. AI runs through the definitions and four offence provisions, and every one is a prohibition on conduct. Nothing here asks a developer or deployer to do anything, so there is no duty a monitor could observe.
Whoever, during the period from publication of notification under Article 11 to the date of publication of result of the election in the official gazette knowingly creates, publishes, shares, or promotes any false or misleading information, image, video, audio, or other content-whether created manually or generated by artificial intelligence (AI) tools- with intent to harm the reputation of any contesting candidate or to influence the election result or otherwise impair the electoral integrity and environment, or uses automated bots, fake accounts, or synthetic media to disseminate false information against opposition candidates, political parties or the Commission, shall be guilty of corrupt practice under Article 73 and shall be punished thereunder.
The Order is an English-language instrument, so this is the operative text itself. Punctuation as printed.
liability of the individual and the political party, candidate, campaign organisation or media agency on whose behalf the offence is committed
Reaches the party or agency behind the poster.
Citation caution. Article 73A was first introduced by section 27 of Ordinance No. 56 of 2025, and that Ordinance was repealed. The Article survives in the consolidated Order, attributed to section 29 of Act No. 03 of 2026. Cite the Act, never the Ordinance.
Analysis. Written for AI-generated election content and enforceable at two to seven years. It bites only inside the election window, requires intent, and imposes no duty to disclose or label. It was in force for the February 2026 election and no enforcement under it has been traced.
AI and Generative AI
The hook that places AI services inside the regulator's licensing jurisdiction.
an AI-based integrated regulatory framework
A mandate to build a framework, not a framework.
বাল্ক কৃত্রিম বুদ্ধিমত্তা কল প্রতারণাReaches economic harm from automated AI calling.
Analysis. The clearest in-force statutory recognition of AI in Bangladeshi law, and it confers jurisdiction rather than obligation. No regulation, directive or licence condition has been issued under any of it.
deployment of AI applications on and connected to the platform for administrative simplification and anti-corruption
A permission for government AI use. It binds no one.
one of six expert seats reserved for an artificial intelligence or large language model expert
The only place AI expertise is written into an institution.
establishment of the Authority, executive chairman and six members
The regulator the data protection Act empowers, and not the body the draft AI policy names.
Analysis. The institutional layer. It creates the authority that would host any future AI supervision, and authorises the state to use AI on its own platform without regulating anyone else's.
in force from 6 November 2025, except section 23 and sections 31 to 35, which commence on a date notified after eighteen months from issuance
So the enforcement layer is not live before about 6 May 2027.
Searched for the Bengali terms for automated, profile and algorithm. The only hit defines encryption.
Analysis. For AI purposes the significance is what it leaves out. The automated-decision layer that carries most of the AI weight in comparable statutes, the article 22 analogue, is simply absent.
deleted by section 61 of the Digital Security Act 2018
The bdlaws footnotes record each deletion. Section 57 is gone.
constitution of the Cyber Tribunal and the Cyber Appellate Tribunal
Both 2026 statutes route their adjudication through these.
Analysis. Mostly hollowed out, but structurally load-bearing: the 2026 statutes borrow its tribunals rather than building their own.
where right to information matters are concerned, the Right to Information Act prevails notwithstanding the overriding effect of the cyber security legislation
Verified by cross-reference from the cyber statute; the 2009 Act itself was not opened.
Analysis. Carries no AI provision. It earns its place because the cyber statute, which does reach AI, expressly yields to it.
The Commission portal returned HTTP 403 twice and the gazette text could not be retrieved. Reported: a ban on malicious AI use in campaigning, plus social-media registration for candidates. No rule number is stated because none could be verified.
Analysis. In-force subordinate legislation that is genuinely AI-specific, and the weakest-evidenced instrument on this page. Listed because omitting it would understate the law, flagged because its text is unverified.
Technology-neutral, adopted before generative AI, and the most recent adopted policy in the field.
Analysis. Listed to make a negative visible: there is no adopted national AI policy or strategy of any year, and this is what sits in its place.
Platforms must implement: AI-generated content labeling
The only labeling mandate in the Bangladeshi corpus, and it binds no one.
a right to counterfactual explanation, expressly not disclosure of proprietary algorithms, and a right to human review of automated decisions with legal or significant effect
Would supply the automated-decision layer the enacted data statute omits.
initiate the drafting of a comprehensive Artificial Intelligence Act by 2028
The draft contemplates its own replacement by statute.
Citation caution. The draft names a National Data Governance Authority as central AI regulator. No such body exists; the statute enacted two months later created the National Data Management Authority instead. The naming is unreconciled.
Analysis. Consultation closed four days before the general election and no adoption step has followed in roughly six months under a changed government. Everything that would create an operator duty in Bangladesh is in this document, and none of it binds.
Observer, announced 1 August 2026. Source: Ministry of Foreign Affairs statement carried by the state news agency; no primary text published.
Party by adoption, 23 November 2021. Source: UNESCO legal-affairs record.
Completed and published; launched 4 December 2025. Source: UNESCO Bangladesh country page and the published report.
Adopted by consensus 22 September 2024. Source: A/RES/79/1 annex.
Consensus adoption 21 March 2024; Bangladesh a co-sponsor. Source: A/78/PV.63 verbatim record.
Consensus adoption 1 July 2024; Bangladesh a co-sponsor. Source: A/78/PV.97 verbatim record.
Member state since September 2023. Source: Organization member record.
Axes: specificity (x, 0-100) is how squarely the instrument is written for AI; force (y, 0-100) is how hard it binds. Placements are editorial judgments for the comparison plot; the tier column is the authoritative classification. The pattern is a country that leads with AI-specific policy at the bottom of the force axis, above a binding tier of general law that carries no AI specificity.
Tap a marker
The binding and prescriptive corner is the crowded one: the CAC-led measures put force and detail together, service by service. The sector guidance sits just left of the binding line, nonbinding in form but treated as obligatory. Dashed markers are drafts; the National AI Law would consolidate the whole stack into one statute.
No adjudicated court case where artificial intelligence is materially at issue was found in Bangladesh, and no dedicated deepfake offence with a citable article exists. Bangladesh has essentially no public court-judgment database, so a non-public matter cannot be ruled out, but nothing is reported. Two matters are recorded below as reported, neither an adjudicated court case, and per the fabrication screen no case number is invented.
Nothing has been adjudicated. The Supreme Court judgments index returns no record for artificial intelligence, deepfake or ChatGPT, though that portal offers no full-text search of judgment texts, so the negative rests also on the prosecutor's statement of August 2025. One matter carries a published case number and it is reproduced exactly; every other matter is recorded as having none, and none is invented.
Bangladesh replaced both of its intellectual-property statutes within the last three years, and both anchor authorship and inventorship to a person, while the copyright exceptions have been delegated to rules rather than enumerated in the Act.
The AI-IP Index reads a country on two questions that usually get folded together. Accommodation asks how far the law will go to protect intellectual property that a machine helped produce. Institution asks how much administrative machinery sits behind that law: its guidance, its international submissions, its disclosure duties.
A country can be generous on paper and have built almost nothing, or cautious and highly organised, so the two are scored apart and reported as a pair. Each rests on three questions, scored 0, 0.5 or 1.
How far the law will protect intellectual property that a machine helped make.
How much administrative machinery sits behind that law.
The scale runs the same way on both axes: 0 is the open or developed end, 1 is the restrictive or absent one, and 0.5 sits between for anything conditional or untested. Read as a pair, the two scores show whether AI-assisted work can be protected here, and whether the office has said so in writing.
Within each axis the three questions carry equal weight. That is a deliberate choice, not an oversight. An AI developer might reasonably value freedom to train above patent inventorship, but weighting is where indices lose their credibility, so the score stays a flat mean and the choice is stated in the open. Each cell also cites a statute section, a case or an office page, so any single score can be challenged without disputing the rest.
Two positions are worth naming. A country can be open on the law but thin on machinery, an open-but-undeveloped stance that bets on ambiguity. Or it can be restrictive on rights yet run an active office, a restrictive-but-developed stance that reflects a settled policy choice rather than a gap. The pair of scores tells those two apart where a single ranking would blur them.
A rubric is only as good as the results it produces on cases where the answer is already known. These three jurisdictions sit outside this series and are scored with the same six questions, as a check. Their law is the best documented anywhere, so if the scale puts them where the law actually sits, the scale is doing its job. They are reference points, not editions.
Firm judicial settlement at both poles and the weakest statutory position on training. The courts have closed the door on AI inventors and AI authors, while everything on training rides on unresolved fair-use litigation. There is no per se duty to declare AI use in a filing, only a candour duty that bites where the use is material.
Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022); Thaler v. Perlmutter (D.C. Cir. 2025), cert. denied 2 Mar 2026; Bartz v. Anthropic (N.D. Cal. 2025); USPTO revised inventorship guidance, 90 Fed. Reg. 54636 (28 Nov 2025); 89 Fed. Reg. 58128 (17 Jul 2024); 89 Fed. Reg. 25609 (11 Apr 2024).
The most permissive statutory position on training and the strongest transparency duty in force, sitting over an authorship question no court has answered. Read the label carefully: training and transparency run through EU law, but patents run through the EPO under the European Patent Convention, which is not an EU institution and covers 39 states.
Directive (EU) 2019/790, arts. 3 and 4; Regulation (EU) 2024/1689, art. 53(1)(c) and (d); EPO J 8/20 (2021); EPO Guidelines G-II 3.3.1 (2026 edition); Infopaq C-5/08; Painer C-145/10.
Liberal on patents since the Supreme Court abandoned the Aerotel test in February 2026, conservative on copyright, and currently thin on AI-specific machinery after the dedicated AI examination guidance was withdrawn. Section 9(3) nominally protects a work with no human author for 50 years, the widest AI-output right on paper, but no court has ever applied it to generative AI.
CDPA 1988 ss. 9(3), 12(7), 29A, 178; Thaler v Comptroller-General [2023] UKSC 49; Comptroller-General v Emotional Perception AI [2026] UKSC 3; UKIPO practice notice, 14 Jul 2026; Data (Use and Access) Act 2025 ss. 135 to 137.
The check earned its keep twice. The untested rule held: section 9(3) of the UK Copyright Act reads like the widest AI-output right anywhere, yet because no court has applied it to generative AI it scores as untested rather than open, which is the correct answer and the one a headline reading would miss. The check also exposed a limit worth stating. The Institution axis measures AI-specific machinery, not general office capacity, so a long-established office can score poorly when it has withdrawn its AI guidance and imposes no disclosure duty. Read that axis as what an office has built for AI, nothing wider.
Each row below carries the test we applied and the finding behind the score, with a source. The rationale, not the number, is the point.
Scores verified 4 August 2026, against AIP Index codebook v1.0. A score is only true as of its date.
What we look forA granted patent or a clearly workable examination route, not merely a silent statute.
The findingThe Patent Act 2023 contains no definition of inventor at all. Section 2 defines twenty-one terms and omits that one, so inventorship is established only indirectly: section 4(a) admits an application by a person claiming to be the proper and first inventor, and section 8(2) requires the name of the true and first claimant with a declaration of belief. Section 2(14) defines person to include natural persons, Government, companies and associations, so nothing on the face of the statute expressly confines inventorship to a human, and nothing expressly opens it to a machine. Section 6(1)(b) separately excludes business methods and any computer programs or algorithms relating to them. No DABUS-type filing has tested Bangladesh and no court has ruled.
What we look forProtection applied where human choices shape the work, and refusal of purely machine output.
The findingThe Copyright Act 2023 anchors authorship to a person in every relevant limb of section 2(23), and section 2(31) defines person to include companies and associations. The decisive point is a subtraction: the repealed Copyright Act 2000 contained an express computer-generated-works regime, giving authorship of works created by computer to the person or institution who created them together with a dedicated sixty-year term, and the 2023 Act dropped that framing while keeping authorship tied to a person. Purely machine-made output therefore has no qualifying author and no authorless category exists, although a corporate deployer could in principle be the creating person. Untested by any court.
What we look forA statutory or judicial mining lane. Silence scores against the miner, since the reserved-rights default governs.
The findingThere is no training or text-and-data-mining exception; the operative text was searched for mining, training and machine terms and returned nothing. The structure is worth stating precisely, because it is not the usual closed list: section 70(1) provides that reproduction, adaptation, recording, broadcast, dissemination, display, publication, fair use and translation do not infringe where done in accordance with purposes and conditions specified in rules made for the purpose, with a separate accessible-format carve-out. The scope of fair dealing therefore now sits in subordinate legislation rather than in the Act. Either way there is no mining lane, so the reserved-rights default governs the use of works for training.
What we look forA traceable, substantive submission on the record, not bare membership.
The findingBangladesh is a WIPO member and no submission to the WIPO Conversation on intellectual property and frontier technologies was located. The negative is recorded as qualified rather than confirmed, because both of the archive's submission-search endpoints returned the same eighty-two records of the separate index of AI initiatives in IP offices rather than any submission roster, so the roster could not be enumerated. A WIPO intellectual-property management clinic in Bangladesh concluded in 2026, which is technical assistance rather than a policy submission. Membership without a submission earns the middle.
What we look forA standalone AI examination text, not AI handled quietly under general practice.
The findingThe Department of Patents, Designs and Trade Marks has published no guidance on examining AI-related filings. Verified on two independent checks: the department's laws, notifications and circulars, publications, notices and legislative-information pages were retrieved and scanned on 4 August 2026 with no occurrence of artificial intelligence in either language, and the WIPO index of AI initiatives in IP offices was enumerated in full, all eighty-two records across twenty-six countries and territories, with Bangladesh absent.
What we look forA mandatory declaration inside the filing, not a proposal or an informal request.
The findingNo filing requires declaring the use of artificial intelligence, and no in-force instrument imposes a labeling, provenance or transparency duty on AI-generated content. This is a considered absent rather than a middle, because Bangladesh does have binding AI-specific law and it is entirely prohibitory: article 73A of the Representation of the People Order criminalises AI-generated election disinformation, and section 25 of the Cyber Security Act 2026 criminalises AI-created or AI-edited sexual imagery, but neither requires anyone to disclose or mark anything. The one labeling mandate in the corpus sits in an unadopted draft policy.
Method. Six questions, three per axis, scored 0, 0.5 or 1 against AIP Index codebook v1.0. The three questions in each axis are weighted equally by choice, and the two axes are reported as a pair, never blended into one number. Every score cites a statute, case or office page, carries a verified date, and records its direction of travel apart from the level. Positions are editorial judgements read from primary law and office practice, not official scores. WIPO's public catalogue of IP-office AI initiatives does not score jurisdictions on this basis, so the readings here come straight from the source law. Not legal advice.
Bangladesh governs artificial intelligence through policy, not law. Two national AI policies are in effect, a National Artificial Intelligence Policy of August 2024 and a General Policy for the Safe and Ethical Use of Artificial Intelligence Systems of April 2025, and both are deliberately non-binding: the first states in terms that it will not impose concrete obligations on the providers or deployers of AI systems. What binds is general law: a Personal Data Protection Law fully in force since February 2026, a new Information Technology Crimes Law of 2026, the electronic-transactions and telecommunications statutes, and the intellectual-property laws, which confine authorship and inventorship to human makers. Where Bangladesh stands apart from its sparse peers is its international posture. It is one of the twenty-nine founding member states of the World Artificial Intelligence Cooperation Organization, and unlike most of the founding roster that membership is confirmed in Bangladesh's own official record, named signatory and all. For a jurisdiction with no binding AI law of its own, that confirmed founding seat is the profile's centre of gravity.
The site does three things: it maps every instrument that governs AI in Bangladesh, 14 of them, ordered by how hard each one binds; it reads the same instruments a second time by how squarely each is written for AI; and it records the matters where AI meets the law, with what was searched and found empty stated as plainly as what was found. Every claim traces to a primary source, and absences are recorded rather than papered over.
A membership claim enters this table only from an official list: a declaration annex, a treaty signature table, an organization's own record or an official government statement. Bangladesh's standout credential is a founding seat in the AI cooperation organization that is confirmed in its own official record, named signatory and all, rather than merely reported.
Know of a ruling or proceeding in Bangladesh where AI is central? Send a source link; it will be reviewed before publishing.
Submit a case