Connect with us

AI

Malaysia’s AI Copyright Draft Points at Singapore and Japan

MyIPO’s closed Act 332 paper cites Singapore and Japan for AI training, while proposing a 12-hour takedown clock and live pirate-site blocks.

Published

on

MyIPO closed a 3 July to 14 August 2026 public comment window on Copyright Act 1987, known as Act 332 changes that name Singapore and Japan as models for AI training. Creators, publishers and media groups, in the agency’s own words, want to be asked first, paid fairly, and told how their works are used. The same paper would put online platforms on a proposed 12-hour takedown clock and let courts extend pirate-site blocks to mirror domains without a new lawsuit.

That mix is the draft. It is not a statute, and no bill has been tabled. The comment window is already shut.

The Consultation Paper Points at Singapore and Japan

MyIPO, which sits under the Domestic Trade and Cost of Living Ministry, put the package out as a Regulatory Impact Analysis exercise. In a 10 July 2026 briefing on the paper, the agency said it wants a legal frame for AI use of copyrighted works that stresses transparency, fairness and appropriate compensation. It also said Malaysia is weighing a design that protects creators while still supporting AI work.

Creators, publishers and media want to be asked first, paid fairly and told clearly how their works are being used.

MyIPO, consultation briefing, July 2026

Counsel who read the paper in July described a different reference set for the training piece. Raymond Tan, a Kuala Lumpur intellectual property partner, wrote in a 14 July 2026 client note that the proposals cite Singapore and Japan, and treat analysis of works for AI training (finding patterns, correlations, insights) as a different act from uses that draw value from human enjoyment. That is the split those two countries already write into statute, and it is not the same thing as a licence for every ingested file.

The paper also says AI-assisted or AI-generated output would still need enough human authorship and creative judgement to qualify for copyright, matching the originality test in Section 7(3)(a) and the definition of an author in Section 3. Pure machine output would stay outside the right, on that reading.

Fasyarini Azahari, a Dewan Negara member, told the house on 7 August 2026 that the law has to keep evolving as AI changes the future, and she tied that job to KPDN through MyIPO. The comment form closed a week later. There has been no public MyIPO account of how many submissions arrived, or which way the training clause will be written into a bill.

Platforms Would Face a 12-Hour Takedown Clock

The AI heading is what travelled. The operational load sits on service providers, and it is heavier than the current notice rules.

THE ENFORCEMENT TOOLS IN THE DRAFT

  • Takedown clock: Safe-harbour rules under Sections 43C to 43E would tighten, with a proposed Section 43H notice-and-takedown window that could run to 12 hours, against the 48-hour disable-access period that applies now after a valid notice.
  • Live site blocks: Courts could grant dynamic injunctions under an expanded Section 37(1)(a), covering not only named pirate sites but mirror, clone and successor domains without a fresh suit each time.
  • Domain seizure: Criminal tools would add seizure and forfeiture of domain names used in copyright infringement, on top of the usual site-blocking orders.

Tan flagged the obvious follow-on: a 12-hour clock and live blocks help rights holders, and they also raise questions of scope, proportion and due process that the consultation paper did not fully answer. Providers that want to keep safe harbour would also face stricter conditions once they have actual knowledge of infringing activity.

Technological protection measures and rights-management information under Sections 36A and 36B would be reinforced as well, with liability extended to devices, services and activities designed to strip those controls. For a platform, that is a product and process problem, not a press-release problem.

A Hearing No Licensing Body Can Block

Royalty fights are still gated. Under the Act as it stands, the Copyright Tribunal sitting in Petaling Jaya can hear a royalty dispute under Section 59C(1) only if both the licensing body and its member consent. Either side can stop a case by refusing. MyIPO proposes to let either party refer a dispute on its own, and it says all cases would be resolved within 60 days after hearings end.

That change lands on collective management organisations declared under Section 27A, which already sit in a thin spot when a declaration lapses. The paper would tighten CMO governance, royalty handling and licensing practice, and it would fill the gap that appears when a Section 27A declaration is not renewed, including what happens to existing licences and unpaid royalties.

That gap is not theoretical. The Controller of Copyright told Music Rights (Sarawak) Berhad that its CMO declaration would not be renewed, and that from 18 March 2026 the body could not go on licensing as a collective under Act 332. A faster tribunal with no mutual-consent lock would give members a route that does not depend on the collection body agreeing to be sued in front of it.

Tan’s note warned CMOs to expect more tariff and distribution fights once the consent bar falls. The tribunal’s own docket is still short; one published decision, Abdul Karim Al Ali v Music Authors’ Copyright Protection Berhad (CTMR2024001), was issued on 20 August 2025.

Orphan Works and a Cut of the Resale Price

Two new property rights sit beside the AI clause, and both would move money and access in the other direction from a training exception.

Orphan works (books, newspapers, photographs, films whose owners cannot be found) have no dedicated frame in the Act now. After a diligent search failed, the draft would allow use for education, research, preservation or public-interest projects. MyIPO said libraries and universities could digitise older holdings, and that the extra access would feed documentaries, exhibitions and other new work. The paper still has to define how diligent is diligent, who keeps the search file, and what happens if an owner later appears.

The Artist’s Resale Right is also new. Malaysia does not now give artists a cut when a work is sold again, so later gains stay with collectors and galleries. MyIPO said the right would cover fine art, photography, illustrations, crafts, designs, architecture and prints, and that over 90 countries already run similar schemes. Southeast Asia has not. Auction houses and dealers would become the collection point, and the rate, the threshold and the collecting body are still open. MyIPO tied the idea to fairness and to the creative industries’ contribution to GDP, without publishing a tariff.

A quieter boundary fight sits in Section 13B, where copyright for paintings, drawings and digital art meets industrial-design protection for goods made by an industrial process. The draft would let the minister spell out when an article is made by industrial means, and carve out works that are mainly artistic or literary. KPDN wants fewer fights over which file to open when a product shows up in the market.

Taken together, those planks match the five Copyright Act 1987 reforms MyIPO listed for comment, with extra machinery (CMO rules, voluntary-notification evidence, the 12-hour clock) sitting underneath the AI headline.

What Singapore’s Section 244 Lets Trainers Copy

If Malaysia copies the statutes it named, trainers get a permitted-use path. If it copies the ask-first-pay language from its own briefing, they do not. The paper has not chosen in public.

Singapore’s Copyright Act 2021 treats copying for computational data analysis as a permitted use when the copier has lawful access, uses the copy only for that analysis or to prepare for it, and does not pass the copy around except to check results or for collaborative research. Circumventing a paywall is not lawful access. An illustration in the Act is using images to train a program to recognise images. Contract terms cannot switch that exception off. The computational-analysis division commenced on 21 November 2021.

Japan’s 2018 non-enjoyment exception (Article 30-4) goes further on paper: a work may be used, to the extent needed, when the point is not to enjoy the thoughts or sentiments in it, including data analysis and recording works as AI training data, without the owner’s permission. It drops away if the use would unreasonably prejudice the copyright owner.

Europe is the other pole. The Digital Single Market Directive lets commercial text-and-data mining go ahead unless the owner reserves rights in a machine-readable way, and the AI Act has, since 2 August 2025, required general-purpose model providers to have a policy for finding and honouring those reservations. The United Kingdom, after a consultation that drew more than 11,500 responses, said on 18 March 2026 that it would not add a broad commercial training exception, so licences remain the default unless an existing exception fits.

HOW FOUR SYSTEMS TREAT TRAINING COPIES

Place Training on lawfully accessed works What a rights holder can still do
Singapore Permitted computational data analysis (Section 244) Cannot contract out of the exception; paywalls still count
Japan Permitted when the use is not for enjoying the work Exception fails if the use unreasonably hurts the owner
European Union Commercial mining allowed unless a machine-readable opt-out is set Opt-out plus an AI Act duty to honour it
Malaysia draft Paper cites Singapore and Japan; briefing also talks of pay and notice No bill yet, so neither path is locked

An August 2026 Oxford Economics seven-country study of AI training rules, prepared for Google, still slotted Malaysia as “uncertain (nascent discourse)” with no fair-use clause and no clear text-and-data-mining exception, then noted in a July addendum that the MyIPO paper points at Singapore and Japan. The same study argued that copyright flexibility is now a location issue for frontier AI, which is a Google-funded conclusion and should be read as such. The local fact underneath it does not depend on that sponsor: Malaysia has not yet said, in statutory text, which of those four columns it wants.

Compute Arrived Before the Statute Did

The National AI Action Plan 2026-2030 took effect on 28 July 2026, while the copyright form was still open, with Digital Minister Gobind Singh Deo still using the “AI Nation by 2030” line. The comment window then closed. The compute deals did not wait.

THE DATES AROUND THE CLOSED PAPER

  1. 3 July 2026: MyIPO opens public comment on proposed Act 332 amendments, including AI training, piracy blocks, orphan works, resale royalties and tribunal rules.
  2. 28 July 2026: The National AI Action Plan 2026-2030 takes effect, with AI Malaysia Berhad and a safety institute stood up in Cyberjaya.
  3. 14 August 2026: The comment window closes. No draft bill follows in public.
  4. 8 September 2026: OpenAI signs a multi-year deal with Firmus to take dedicated capacity at two planned Malaysian data centres, as an anchor customer on Firmus’s AI Factory platform.
  5. 14 September 2026: Investment, Trade and Industry Minister Johari Ghani says the Data Centre Task Force has had more than 40 applications this year to set up or expand sites.

Johari, speaking at Data Centre Nexus 2026, said Malaysia approved RM218.5 billion in investments across sectors in the first half of 2026. He put the services share at RM149.6 billion, or 68.5 percent, and said data-centre and cloud projects alone accounted for RM95.8 billion, close to 44 percent of all approved investment in the period. Data-centre-related approvals, he said, have reached RM385.7 billion since 2021. He wants the next phase to show up as AI use, local products and jobs, not only halls of servers.

Those halls can be filled under today’s Act. Training a model on Malaysian news, music, books or film still sits in the gap the closed paper was meant to fill. Publishers asked for notice and pay. The models named in the paper let a trainer who already has lawful access copy for analysis without a new licence. Platforms, meanwhile, would have to rebuild takedown desks around a 12-hour clock and keep maps of pirate mirrors current under a single court order.

On 14 September 2026 the task force was still counting new data-centre filings. The copyright text those operators would train and host under remains a closed consultation, not an Act of Parliament.

Disclaimer: This article is news reporting on a closed public consultation and related policy statements, and it is for information only. It is not legal advice and it does not tell rights holders, platforms, collective bodies or AI developers how to comply with the Copyright Act 1987 or with any future amendment. Anyone who needs to act on licensing, takedown duties, resale royalties or training-data use should consult a Malaysian intellectual property lawyer before changing contracts or systems. Figures, section numbers and proposal details reflect MyIPO materials and public statements from the 2026 comment window and may change before a bill is tabled or passed.

Harry is the editor of Oton Technology, an independent site he owns and edits, covering the part of technology that people actually have to act on. After ten years in journalism, first reporting and then editing, he works from primary material by habit: the advisory rather than the write up of it, the filing rather than the press release, the changelog rather than the launch video. Every figure in an article carries its source and its date, and where a number comes from a vendor or an analyst model rather than a count, he says so plainly instead of letting it stand as established fact. What he leaves out is anything he could not verify himself, which on a beat full of unnamed supply chain claims removes a great deal. That standard applies across all the sections the site publishes for an international audience, from artificial intelligence and security to phones, computers, gaming, crypto and the software businesses depend on. He corrects errors in the open and labels them, because a site that hides its mistakes is asking readers to trust the rest on nothing.

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Trending