Malaysia

The Federal Constitution is silent on AI. However, the Part II fundamental liberties are increasingly engaged by AI development, deployment and governance. The relevant provisions are summarised below:

Policymakers have begun to acknowledge the challenges posed by AI-generated content. In 2025, the Minister in the Prime Minister’s Department (Law and Institutional Reform) noted that courts may need to be prepared to address the evidential risks associated with AI technologies, particularly where fabricated visuals, audio or documents could affect the integrity of proceedings. While Malaysian courts have long dealt with issues of digital evidence and authenticity, there has been no definitive constitutional or procedural ruling specifically addressing AI-generated evidence.

The Federal Court in The Government of Malaysia v Heidy Quah Gaik Li (Civil Appeal No. 08(f)-35-02/2024(W)) unanimously allowed the government’s appeal in part, reversing the Court of Appeal and holding that the words “offensive” and “annoy” in section 233(1)(a) of the Communications and Multimedia Act 1998 (CMA) are not unconstitutional. The case is the leading recent authority on the scope and limits of Article 10(1)(a) of the Federal Constitution tested against the criminalisation of online communications. The court held the impugned words target communications that fall outside protected speech, consistent with the CMA’s purpose of protecting and safeguarding users of network communications. 

The reasoning is directly relevant to AI-generated offensive content disseminated online.

The Patents Act 1983 defines an “inventor” which suggests that inventors must be natural persons. This is supported by section 18, which assigns the right to apply for a patent to the inventor. The Patents Regulations 1986 require patent applications to include either the personal identification of inventors or signed written declarations when anonymity is sought.

AI cannot be named as an inventor; AI-assisted inventions remain patentable where a natural-person inventor made an inventive contribution, but inventions autonomously generated by AI are unlikely to be patentable.

Users of AI must ensure that AI-generated works (outputs) do not infringe third party rights, as the use of infringing materials may nevertheless entail liabilities under the relevant intellectual property laws, regardless of the fact that such materials were used and/or generated by AI. Unlike the UK Copyright, Designs and Patents Act 1988 (CDPA), which explicitly recognises copyright protection for computer-generated works, the Malaysian Copyright Act 1987 has no provision for computer-generated works. While “author” under section 3 of the Copyright Act 1987 may include a body corporate, this interpretation conflicts with other copyright provisions, such as:

• Duration. Protection lasts for 50 years after the author’s death, inherently implying that an author must be a natural person.
• Creative input requirement. Copyright law typically requires human authorship, where an individual expends a certain level of skill, judge­ment or effort in creating a work. Further, the human element input here arguably is confined to the input of ideas or concepts; mere prompts, input, descriptions or instructions by the user are unlikely to meet the threshold and may not themselves be protectable. Given these factors, it is likely that AI-generated works may not qualify for copyright protection in Malaysia. Without clear statutory provisions, businesses and creators leveraging AI should explore alternative protection strategies, such as contractual allocation of ownership in AI outputs, trademark protection for AI-created branding elements, and trade secret protection for AI algorithms and proprietary content.

Malaysia has no trade secrets statute. Protection rests on the tort of breach of confidence. To qualify, information must have the necessary quality of confidence, must have been imparted in circumstances importing an obligation of confidence, and there must be an unauthorised use of the confidential information to the detriment of the party communicating it.

Practical points for AI:

• Model weights, training datasets, prompts, fine-tuning data and architecture can be protected as confidential information/trade secrets, provided reasonable secrecy measures are in place (access controls, NDAs, need-to-know).
• Inputting confidential data into third-party AI tools may destroy confidentiality if the provider can reuse inputs for training; enterprise terms should prohibit reuse and require segregation.
• Employee and contractor agreements should expressly cover AI-related confidential information and post-termination obligations.

There is no reported Malaysian decision yet on AI training data or AI-output infringement. Businesses should expect the AI Governance Bill to clarify the position.

The PDPA governs processing of personal data in commercial transactions, including those involving Automated Decision-Making and Profiling (ADMP). The Personal Data Protection (Amendment) Act 2024 introduced, in phases from 2025:

• mandatory breach notification;
• a Data Protection Officer (DPO) appointment obligation;
• direct obligations on data processors;
• a revised cross-border transfer regime; and
• increased penalties.

The JPDP Guidelines on DPIA, Data Protection by Design and ADMP (launched 30 April 2026) are central to AI:

• Training data. Using personal data to train AI is “processing” and requires a lawful basis, purpose limitation and (generally) consent or a recognised exemption.
• Data Protection Impact Assessments. DPIAs are expected where the processing involves technology that is new to the organisation and where the processing involves the implementation of technology that is new to the organisation’s operational environment.
• Automated decision-making. Data subjects must be informed of the use of AI in the processing of personal data involving ADMP when providing a written notice (personal data protection notice or privacy notice) regarding such processing. The data controller may consider the appropriate scope of the explanation on the use of AI in such notice, ensuring that it is not excessively lengthy or overly technical for the data subject.

Personal data refers to any information in respect of commercial transactions, which is:

• being processed wholly or partly by means of equipment operating automatically in response to instructions given for that purpose;
• recorded with the intention that it should wholly or partly be processed by means of such equipment; and
• recorded as part of a relevant filing system or with the intention that it should form part of a relevant filing system, that relates directly or indirectly to a data subject, who is identified or identifiable from that information or from that and other information in the possession of a data controller.

Given these parameters, businesses utilising AI to analyse customer data, automate decision-making or personalise user experiences must ensure that their AI systems are designed in compliance with the PDPA. The PDPA establishes key compliance obligations relating to data collection, processing and storage, ensuring that businesses employing AI-driven analytics or automated decision-making maintain strict data protection standards. AI technologies often process vast amounts of personal data, making compliance with the PDPA essential for businesses leveraging AI-driven solutions.

To regulate data flows and AI driven data-sharing arrangement, the PDPA requires, among other things, compliance with seven Personal Data Protection Principles, which are set out in sections 6 to 12 of the PDPA. Data controllers are expected to comply with all seven principles, which are summarised as follows:

• General principle. Any processing of personal data generally requires the consent of the data subject, and only in relation to the purpose of processing for which consent was given. 
• Notice and choice principle. The data subject shall be notified by written notice of the purpose for which data is collected and processed, and such other information as prescribed by section 7(1) of the PDPA (“Privacy Notice”). The Privacy Notice is a written statement explaining how the data user processes the personal data of the data subject. By examining the Privacy Notice, the data subject should get a clear picture of how the data user will process the data subject’s personal data and what options are available to the data subject. 
• Disclosure principle. Generally, no personal data shall be disclosed to any third party without the consent of the data subject.
• Security principle. Practical steps are required to be taken to protect the personal data from any loss, misuse, modification, unauthorised or accidental access or disclosure, alteration or destruction.
• Retention principle. Personal data shall not be kept longer than is necessary for the fulfilment of the original purpose that it was obtained for.
• Data integrity principle. This is the obligation to take reasonable steps to ensure that personal data is accurate, complete, not misleading, and kept up to date.
• Access principle. A data subject shall be given access to their personal data and be able to make corrections where it is inaccurate, incomplete, misleading or not up to date. 

There is no general open-data statute binding the private sector; government open-data initiatives are policy-driven. The principal data-sharing considerations are:

• IoT. The assessment must address IoT-specific features — notably reliance on decentralised edge processing — and what the PDPA requires in that context, including whether such processing is necessary and not excessive.
• Healthcare IoT. Given the prevalence of IoT in healthcare, deployments must also meet requirements for medical data and records. The PDPA treats information on a data subject’s physical or mental health as sensitive personal data, the processing of which generally requires the data subject’s explicit consent (not mere consent). IoT use in healthcare may therefore require patients’ explicit consent for processing their health data for the relevant purposes. Compatibility with medical record-keeping and retention requirements — for example, under the Private Healthcare Facilities and Services Act 1998 and the Private Healthcare Facilities and Services (Private Hospitals and Other Private Healthcare Facilities) Regulations 2006 — must also be considered.
• Public-sector data sharing. The Data Sharing Act 2025 primarily governs the sharing of data held by a public sector agency with other public sector agencies in Malaysia. A “public sector agency” means the agencies in charge of the public services under Article 132(1) of the Federal Constitution (excluding the services in paragraphs (f) and (g)) — namely the armed forces, judicial and legal service, general public service of the Federation, police force, and the joint public services under Article 133 — and any statutory authority exercising powers vested by federal law. The Act may indirectly affect private entities: where a public-sector data recipient engages a third party to carry out data migration, integration or analytics using the shared data, the recipient must obtain the data provider’s consent before the third party handles the data, and the third party must handle it in compliance with the Act and the applicable data-security requirements.

Sector-specific guidance is also emerging. On 20 November 2025, the MCMC registered a Technical Code on Internet of Things (IoT) – Privacy Requirements (MCMC MTSFB TC G056:2025), developed by the Malaysian Technical Standards Forum Bhd (MTSFB) under sections 184–185 of the Communications and Multimedia Act 1998. It operationalises the PDPA for IoT systems that process personal data (including biometric, health, geolocation and multimedia data), addressing data-controller/processor responsibilities, consent and notice, data minimisation and retention, security, cross-border transfers and a privacy governance model (including DPO and DPIA expectations). As a registered technical code, it provides industry-standard guidance for IoT deployments but does not displace the PDPA itself.

Biometric data (facial, voice, fingerprint) is personal data under the PDPA. The PDPA treats biometric data as sensitive data warranting heightened safeguards. Facial recognition and voice AI deployments should rest on explicit consent or another lawful basis, undergo a DPIA, apply data minimisation and retention limits, and address accuracy and bias risks. Biometric surveillance also engages Article 5 privacy considerations (see Section 1, above).

Malaysia has no general anti-discrimination statute applicable to the private sector. Article 8 of the Federal Constitution guarantees equality before the law but binds the State, not private actors. Sector rules address aspects of fairness: employment (limited grounds), financial services (responsible lending under Bank Negara Malaysia (BNM) policy), and the National AI Guidelines’ fairness and inclusiveness principles.

In practice, businesses managing AI bias should rely on contractual standards, internal governance, algorithmic impact assessments and audits, pending the AI Governance Bill, which is expected to introduce harm-assessment obligations covering discriminatory outcomes.

The CSA, in force since 2024, establishes duties and powers of the Chief Executive of the National Cyber Security Agency (NACSA) and a framework for the designation and regulation of NCII. NCII means a computer or computer system of which the disruption to or destruction of would have a detrimental impact on the delivery of any service essential to the security, defence, foreign relations, economy, public health, public safety or public order of Malaysia, or on the ability of the Federal Government or any of the State governments to carry out their functions effectively. There are 11 NCII sectors, namely:

• government;
• banking and finance;
• transportation;
• defence and national security;
• information, communication and digital;
• healthcare services;
• water, sewerage and waste management;
• energy;
• agriculture and plantation;
• trade, industry and economy; and
• science, technology and innovation.

NCII entities must:

• conduct a national cyber security baseline self-assessment;
• conduct a cyber security risk assessment at least once a year and a cyber security audit at least once every two years (or such higher frequency as may be directed by the chief executive);
• provide NCII information to sector leads upon request;
• implement the measures, standards and processes as specified in the Code of Practice prepared by its NCII sector lead for the purposes of ensuring the cyber security of its NCII;
• submit risk assessment and audit reports within 30 days of completion of the cyber security risk assessment or audit; and
• notify the chief executive and sector lead of any cyber security incident which has or might have occurred in respect of its NCII.

For AI, where an AI system supports NCII, these obligations cascade to AI vendors contractually. A managed Security Operation Centre (SOC) monitoring service or penetration-testing feature in an AI product may itself require a Cybersecurity Service Provider licence, unless the service relates only to systems located outside Malaysia or is provided between related companies.

The Competition Act 2010 prohibits anti-competitive agreements and the abuse of dominant position in the market. Although untested on AI, the following are live risks:

• algorithmic collusion (pricing algorithms aligning conduct, including tacit coordination via shared tools);
• self-preferencing and exclusionary conduct by dominant digital platforms; and
• exploitative dark patterns.

It is to be noted that the Competition Act does not apply to the commercial activities, specified in the First Schedule of the Competition Act, namely communications and multimedia, energy, upstream petroleum operations and aviation services. Competition issues, including alleged anti-competitive conduct, in relation to these activities fall under the jurisdiction of the regulatory authorities that regulate the relevant activities.

Businesses deploying pricing or recommendation AI should assess collusion and abuse risk.

The Malaysian Competition Commission (MyCC) enforces the Competition Act 2010. The MCMC enforces equivalent provisions for the communications and multimedia sector under the CMA. At present, Malaysia has no competition law regulations directly addressing AI or generative AI (GenAI).

In its Market Review on the Digital Economy under the Competition Act 2010, MyCC examined digital markets, including mobile operating and payment systems, online travel agencies (OTAs), e-commerce and digital platforms and observed the use of GenAI across these industries, signalling growing regulatory attention to GenAI in these sectors.

Malaysian law does not recognise AI agents as legal persons. They cannot hold rights, owe duties or be party to a contract. Practical consequences:

• Contracts. Agreements executed by an AI agent are attributed to the human or corporate principal deploying it, applying ordinary Contracts Act 1950 principles (offer, acceptance, consideration, intention). The principal bears the legal consequences of the agent’s actions.
• Agency. Traditional agency law assumes a legal-person agent; an AI “agent” is a tool, not an agent in law. Authority and liability rest with the principal.
• Liability. Harm caused by an agentic system is allocated to the developer, deployer or user under contract, tort or product-liability principles (see Section 8, below). There is no autonomous AI liability.

No agentic-AI-specific rules exist yet. Sectoral overlays apply where agents operate:

• financial services (BNM Risk Management in Technology (RMiT) treats AI within technology-risk and outsourcing frameworks; Securities Commission Malaysia Guidelines on Technology Risk Management apply to capital market entities and address ethical AI/ML use);
• healthcare (Medical Device Act 2012 where an agent constitutes “software as a medical device”); and
• communications (CMA, section 233 and the Online Safety Act 2025 for content transmitted by agents).

The AI Governance Bill is expected to introduce lifecycle obligations and incident reporting that will bear directly on autonomous agents.

AI-related products and services may be regarded as consumer products and their regulation falls under Malaysia’s existing consumer protection framework, principally the Sale of Goods Act 1957 (SOGA) and the Consumer Protection Act 1999 (CPA). If AI malfunctions or causes harm, liability may arise under these laws.

Consumer Protection Act 1999 (CPA)

The CPA serves as Malaysia’s primary legislation for safeguarding consumer rights. Enforced since 15 November 1999, the CPA sets out mandatory standards for goods and services as prescribed by the Minister responsible for consumer protection. The CPA applies to the supply of goods (that are primarily purchased, used or consumed for personal, domestic or household purposes) to consumers (who acquire or use goods of a kind ordinarily acquired for personal, domestic or household purposes, use or consumption).

The CPA 1999 prohibits misleading and deceptive conduct or false representation (sections 9, 10, 12–16), prescribes safety standards of goods (sections 19–24) and implies guarantees as to title, acceptable quality, fitness for purpose, description, price, repairs and spare parts (sections 31–37) as well as manufacturers’ express guarantees (section 38). Under section 67(1) of the CPA, a product is deemed defective if the safety of the product is not such as a person is generally entitled to expect.

In determining what a person is generally entitled to expect in relation to a product, section 67(2) outlines several factors, including:

“(a) the manner in which, and the purposes for which, the product has been marketed;

(b) the get-up of the product;

(c) the use of any mark in relation to the product;

(d) instructions for or warnings with respect to doing or refraining from doing anything with or in relation to the product;

(e) what may reasonably be expected to be done with, or in relation to, the product; and

(f) the time when the product was supplied by its producer to another person.”

Under Section 68(1) of the CPA, the following persons shall be liable for the damage:

“(a) the producer of the product;

(b) the person who, by putting his name on the product or using a trade mark or other distinguishing mark in relation to the product, has held himself out to be the producer of the product; and

(c) the person who has, in the course of his business, imported the product into Malaysia in order to supply it to another person.”

Therefore, AI developers, manufacturers or importers of the technologies may therefore be liable.

Sale of Goods Act 1957 (SOGA)

Unlike the CPA 1999, the SOGA 1957 governs sale of goods in general. The SOGA 1957 generally incorporates, among others, various implied conditions including that the goods shall correspond with the description (section 15), they are fit for the purpose for which they have been supplied, and they are of merchantable quality under the contracts of sale of goods (section 16). Under section 14 of SOGA, in a contract of sale, unless the circumstances of the contract are such as to show a different intention, there is:

• an implied condition on the part of the seller, that, in the case of a sale, the seller has a right to sell the goods, and that, in the case of an agreement to sell, the seller will have a right to sell the goods at the time when the property is to pass;
• an implied warranty that the buyer shall have and enjoy quiet possession of the goods; and
• an implied warranty that the goods shall be free from any charge or encumbrance in favour of any third party not declared or known to the buyer before or at the time when the contract is made.

A key open question is whether AI software supplied as a service is “goods”; embedded AI in a physical product is more readily treated as goods.

Contract/tort

Under Malaysia’s Contracts Act 1950, a legally binding contract is formed when the essential elements — offer, acceptance, consideration and intention to create legal relations — are satisfied. Applying these principles, AI-generated contracts may be enforceable under the Contracts Act 1950 where these conditions are met and no vitiating factors render the contract void or voidable. However, the evolving nature of AI introduces complex questions regarding contractual intent and liability, particularly in scenarios where AI autonomously enters into agreements.

In tort, liability for AI harm is likely to ultimately rest with the system’s proprietor on negligence, vicarious or strict-liability bases, depending on control and foreseeability. The Trade Descriptions Act 2011 (“TDA 2011”) prohibits false trade descriptions and false or misleading statements, conduct and practices in relation to the supply of goods and services. A “trade description” is defined under section 6(1) of the TDA 2011 as an indication, whether direct or indirect and by any means given, of among other things fitness for purpose, strength, performance, nature, quality, place or date of manufacture, with respect to any goods or parts of goods.

Under section 5(1) of the TDA 2011, “[a]ny person who (a) applies a false trade description to any goods; (b) supplies or offers to supply any goods to which a false trade description is applied; or (c) exposes for supply or has in his possession, custody or control for supply any goods to which a false trade description is applied, commits an offence”. The wide scope of section 5(1) means that both suppliers and manufacturers may be held accountable for false trade description. In particular, manufacturers may face greater liability if they affix or incorporate false trade descriptions onto goods before distribution.

In addition to this, there are prohibitions under the TDA 2011 against any false representation as to the supply or approval of goods (section 16, TDA 2011) and false or misleading advertisement (section 18, TDA 2011) in relation to any goods.

Any body corporate in contravention of such prohibitions may be held liable

Malaysia has no AI-specific strict liability regime. The CPA 1999 Part X product-liability provisions (sections 68–72) impose a form of strict liability on producers, the person who, by putting his name on the product or using a trade mark or other distinguishing mark in relation to the product, holds himself out to be the producer of the product and importers, for damage caused by a defect — subject to defences (e.g. the defect not existing when supplied, or the state-of-scientific and technical knowledge at the relevant time was not such that a producer of products of the same description as the product in question may reasonably be expected to discover the defect). This applies to AI embedded in products.

Pure-software AI faces uncertainty over whether it is a “product”. The forthcoming AI Governance Bill and any future AI liability rules may clarify allocation of strict liability for autonomous systems.

Under the CPA, the Minister may prescribe safety standards and order recalls of unsafe goods; suppliers must comply and may face enforcement for non-compliance. For AI specifically, there is no bespoke recall regime, but practical obligations include issuing security patches and model updates, withdrawing or disabling unsafe AI features, and notifying affected users.

No reported Malaysian AI product-liability decision has emerged. Trends to watch include:

• claims arising from AI hallucinations and erroneous automated decisions;
• allocation of liability across the AI supply chain (model provider, platform, deployer); and
• contractual risk-shifting through indemnities and liability caps.

Litigation is expected to increase as enterprise AI adoption accelerates and the AI Governance Bill introduces statutory duties.

Enacted, AI-relevant. PDPA (Amendment) Act 2024 (phased from 2025); Cyber Security Act 2024; Online Safety Act 2025 (in force 1 January 2026).

Proposed. the AI Governance Bill (expected to be tabled by mid-2026), introducing a risk-classification framework, harm assessment, incident reporting, and developer/deployer accountability across the AI lifecycle; and a new Cybercrime Bill to replace the Computer Crimes Act 1997, addressing AI-enabled fraud and deepfakes.

Next steps. Stakeholder engagement and parliamentary scrutiny; sector-specific AI guidance (notably from BNM and the SC) is expected before or alongside the horizontal Bill.

Strategy is coordinated by NAIO under MyDigital, supported by:

• the National AI Guidelines (2024);
• MY-AI Standards (March 2026, 80+ ISO/IEC standards);
• the AI Technology Action Plan 2026–2030; the National Digital Trust and Data Security Strategy 2026–2030 (expected Q3 2026); and
• a proposed independent Data Commission; and regional leadership via the ASEAN AI Safety Network (operational 2026).

The approach is deliberately phased: standards first, then regulation and compliance, then legislation and enforcement.