On 6 October 2026, at a media forum in Kuala Lumpur, the director-general of the Intellectual Property Corporation of Malaysia (MyIPO), Yusnieza Syarmila Yusoff, said the agency is studying an opt-out mechanism. It would let copyright owners stop their works being used for data mining, including AI training. Bernama reports that she called it a possibility under exploration, not a decision. No amendment has been tabled. The way MyIPO designs the mechanism will matter more than the choice to have one.
What MyIPO actually said
Yusnieza split the AI problem in two. One part is input: how training data is used. The other is output: who owns what an AI system generates. The review also covers which data-mining uses should be permitted, by content type, purpose and volume limits. According to SAYS, MyIPO has held engagement sessions with industry stakeholders on these limits. The review is part of MyIPO's ongoing copyright review.
She also set out the tension in her own words. In Bernama's report she said: "We do not want laws to be overly restrictive to the point of stifling trade, investment, and technological growth." She added that the framework "must be user-friendly, IP-friendly, and balanced." That is the right brief. The risk is in how it gets implemented.
The strongest case for an opt-out
The case for acting is serious. Creators, publishers and musicians say their works are copied at industrial scale to build commercial products that may compete with them, usually without notice or payment. An opt-out gives rights holders a simple control. It also creates a bargaining position: a developer who wants a licensed dataset has to negotiate for it. The US Copyright Office thought the issue serious enough to publish a pre-publication Part 3 report on generative AI training on 9 May 2025. The UK's consultation on copyright and AI, which closed on 25 February 2025, listed giving rights holders more control over whether their works train AI models as a stated objective. The UK page also records that the Department for Science, Innovation and Technology published a report and impact assessment on 19 March 2026. A government that does nothing leaves creators with no remedy. Silence is not neutral.
Why design matters more than direction
The opt-out model has an established reference point. The EU's Directive 2019/790 allows text and data mining of lawfully accessible works unless rights holders reserve their rights, and for online content that reservation has to be made in a machine-readable way. The model is not new. What a statute says in a single sentence can take years to become workable. These are the pressure points MyIPO should settle in the text itself.
- How the opt-out is expressed. If an opt-out can be signalled in any form, such as a website footer, a contract clause or an email, no developer can comply at scale. A standard, machine-readable signal lets small Malaysian AI start-ups comply cheaply and gives rights holders something enforceable. Vague wording favours the largest firms, which can afford legal teams to interpret it.
- Who bears the burden. Placing the burden on developers to honour reservations is reasonable. Requiring them to prove a negative across billions of web pages is not. Liability should attach to ignoring a clear, standard signal, not to missing an ambiguous one.
- Scope by purpose. MyIPO's own framing, which distinguishes content type, purpose and volume, points to a graduated regime. Non-commercial research, search indexing and accessibility tools raise different concerns from building a competing commercial generator. A single blunt rule would treat them all alike.
- Contract override. If the opt-out can be overridden by contract terms or left unenforceable, the reform delivers nothing. Parliament should say clearly how the statutory exception and private terms interact.
Input rules should not be confused with output rules
The input and output split is useful, but the two questions need different tools. Ownership of AI-generated results turns on authorship and originality, and it can wait for evidence. Input rules affect investment decisions now. Developers choose where to train models and where to base research teams according to how clear the rules are. A legal framework that is predictable matters as much to them as one that is permissive.
Courts abroad show how unsettled liability is when statutes are silent. On 16 September 2026 the Ninth Circuit rejected an attempt to turn Section 1202 of the Digital Millennium Copyright Act into a broad source of liability in a case over AI models trained on GitHub code. According to EFF, ordinary infringement and contract claims remain open to the plaintiffs. Parties in Malaysia should not have to litigate novel theories to find out what the law permits. Legislating clearly, and narrowly, is cheaper for everyone.
A proportionate path for Malaysia
Malaysia does not need to pass a sweeping AI copyright law next session. A proportionate approach would proceed in three steps.
- Publish the consultation outcome. MyIPO should release a summary of the stakeholder input and its reasoning, so stakeholders can see which arguments carried weight.
- Legislate a narrow exception with a standardised opt-out. Define permitted data-mining purposes, lawful-access conditions and an interoperable, machine-readable reservation. Leave output ownership for a later review.
- Review on a schedule. Add a statutory review after two or three years, so the rules can be adjusted using Malaysian evidence, such as licensing deals concluded, disputes filed and start-up formation, instead of assumptions imported from other jurisdictions.
The agency has so far signalled caution. Its stated aim of avoiding laws that stifle investment is the correct test, and creators are entitled to hold it to that test as well. An opt-out that rights holders can use and developers can honour would serve both sides. One that is vague or over-broad would serve lawyers.
What to watch
Three signals will show where this is heading: whether MyIPO publishes a draft with the opt-out technically specified, whether it distinguishes commercial from non-commercial uses, and whether the output-ownership question is deferred. Until an amendment is tabled, all of this remains a proposal. The next stage of consultation is the best opportunity for developers, creators and civil society to put forward concrete drafting rather than general positions.