On 5 October 2026, Malaysia's Department of Personal Data Protection (PDP) published Public Consultation Paper No. 1/2026: Artificial Intelligence (AI) and Personal Data Protection Framework. We have confirmed the title, the date and that the paper addresses AI under Act 709. We could not read the paper itself. This article therefore makes no claims about which facial recognition provisions it contains, and the analysis below concerns the legal setting it will land in.
The case for acting now
The argument for tighter rules is strong, and it should be stated fairly. Facial recognition differs from other surveillance tools because a face cannot be changed or left at home. A face template is a permanent identifier that can be matched across every camera in a network. Kuala Lumpur City Hall (DBKL) says all 10,000 of its CCTV cameras now run on its AI-enabled command and control system, which uses facial and vehicle recognition to track suspects. Police say the system has helped resolve 50 criminal cases since 2020. Minister Hannah Yeoh has said the cameras are installed only in public areas. A regulator that waits for a misidentification scandal before writing rules has waited too long.
What the amended PDPA already does
Malaysia has some foundations in place. The Personal Data Protection (Amendment) Act 2024 classifies biometric data as sensitive personal data, defined as data derived from the technical processing of physical, physiological and behavioural characteristics. It also raised the maximum penalty for breaching the data protection principles to RM1 million and/or three years' imprisonment, from RM300,000 and two years. Those are real stakes for a private firm that deploys face recognition at a mall, a gym or an office. The PDP's Act 709 page hosts the statute and its consolidated text.
The gap the paper cannot fill by itself
The difficulty is that the highest-risk deployments are public ones. Section 3(1) of the 2010 Act says it shall not apply to the Federal Government and State Government. Commentators also argue that, with no case law to the contrary, agencies operating as part of government structures fall inside that exemption. We found nothing in the sources we could read indicating that the 2024 amendments changed this, and we have not seen the text of the consultation paper. Whether DBKL, a statutory local authority, counts as "government" under section 3(1) is a question we cannot settle here. But if it does, an AI framework built on the PDPA would bind the retailer using face templates for loyalty programmes and leave the state's city-wide network outside its reach.
That would be a perverse result. The private sector faces consent, security and breach-notification duties for biometric data, while the actor with arrest powers and the largest camera estate faces none of them under this law. The consultation is the right place to ask which public bodies the framework will cover and which will remain outside it.
What proportionate rules would look like
Opponents of facial recognition rules often argue for bans. We do not. A blanket prohibition would throw away real benefits, such as finding missing persons, identifying violent offenders and speeding up investigations. It would also push deployment into informal arrangements with even less oversight. The better course is a short list of enforceable conditions that apply to public and private deployers alike:
- Published purpose limits. Say what the system is for. Locating a named suspect in a serious case is different from tracking everyone who walks past a camera.
- Written, accessible retention rules. The shorter the retention period for non-matching faces, the lower the breach and misuse risk.
- Accuracy testing and error reporting. Deployers should disclose match thresholds and false-match rates by demographic group, and a human must review any match before action is taken.
- Independent audit and a complaint route. Someone outside the deploying agency should be able to check use of the system and hear challenges.
- Transparency about the system. India offers a cautionary example. A Delhi Police response to a Right to Information request, reviewed by MediaNama, contained no reference to biometrics, watchlists, match thresholds, retention, deletion or audits. Citizens cannot contest a system when the documents that govern it do not exist.
These conditions are compatible with innovation. Clear rules lower the legal risk for vendors and local authorities, and they make it easier to build public trust in a technology that Malaysia is evidently committed to using.
What to watch in the consultation
Three questions will determine whether the framework is substantive or symbolic. First, does it extend any obligations to federal, state or local public bodies, by amendment or by a parallel instrument? Second, does it treat face templates as sensitive data subject to a stricter lawful-basis test than ordinary personal data, as the 2024 Act's biometric classification would suggest? Third, does it contain any duty to publish impact assessments before a large deployment goes live? The PDP's earlier consultations on data protection impact assessments and automated decision-making show it is willing to ask these questions. The AI paper is where it can answer them.
Malaysia can have both a modern AI economy and accountable surveillance. The test is whether the framework applies to the deployments that most affect people's daily lives.