Following our previous article on the Public Consultation Paper on AI and Personal Data Protection Framework (“Public Consultation Paper”), the public consultation has now concluded, with the proposed Framework moving into the usual process of reviewing feedback and internal finalisation ahead of its official release.
In this article, we will go deeper into the proposed framework to help organisations prepare for what is to come. While public consultation may lead to amendments and refinements, the underlying structure and core concepts often remain a useful foundation for early preparation. With this in mind, we set out the top 7 key takeaways that companies should consider when implementing AI within their organisations, particularly where personal data will be involved and processed, as will very often be the case.
For transparency, both authors, Ong Johnson and Lo Khai Yi, were appointed by the Personal Data Protection Commissioner (“PDPC”) to the drafting committee for the Public Consultation Paper, but all information and insights shared in this article are drawn exclusively from publicly available materials.
1. Lawfully Obtaining Personal Data Does Not Give a Company Unrestricted Permission to Use It for AI Training
When it comes to processing personal data for AI, we often see companies looking beyond the personal data they hold internally, particularly for the purpose of training their AI systems. Companies may also obtain personal data from third-party sources, publicly available sources or external datasets, and these arrangements can take various forms, including purchasing or licensing datasets from commercial providers, accessing data through research collaborations, or collecting information through web crawling and scraping.
However, Paragraph 6.3 of the Public Consultation Paper makes it clear that, even where personal data is obtained from third-party sources, publicly available sources or external datasets, the data controller remains fully responsible for ensuring that its collection and subsequent processing comply with the PDPA and remain consistent with the purpose for which the data was collected or made available.
This is particularly important as more companies source data to develop, train and improve their AI systems internally. Naturally, access to more relevant, high-quality data can help improve an AI system. However, just because personal data can be accessed online, or has been lawfully purchased or licensed, it does not mean that a company can automatically use it for any AI training purpose without restriction. Ultimately, the company still needs to assess the lawfulness of obtaining and processing the personal data, including whether its intended use is consistent with the original purpose for which the data was collected or made available.
For example, individuals may have agreed to their personal health data being collected and licensed specifically for training AI systems to support cancer drug discovery. A company may then lawfully obtain a licence to that dataset for this purpose. However, if the company subsequently uses the same dataset to train an AI system that profiles individuals’ health risks for insurance underwriting, that subsequent use would then be inconsistent with the purpose for which the personal data was originally collected and made available.
Of course, companies might think that developing an insurance risk model may itself be a legitimate business activity, and even if the dataset comes from a legitimate source, it, however, does not automatically permits the personal data to be repurposed in this way because that subsequent use serves a materially different purpose from the cancer research for which the patients’ data was originally collected and licensed.
Therefore, for companies, it is crucial to examine both the source of the personal data and the scope of its permitted use. Before incorporating an external dataset into an AI project, beyond understanding and verifying the source of personal data, companies should also review the original collection purposes and any restrictions on its subsequent use for AI purposes.
2. Your PDP Notice Should Explain How AI Is Used to Make Decisions, Generate Recommendations or Personalise Services
The second takeaway is that, if a company deploys an AI system that makes or supports decisions affecting data subjects, generates recommendations or predictions concerning them, or provides personalised outputs or services using their personal data, the Data Protection Notice (“PDP Notice”) should be updated to properly reflect this.
Paragraph 7.3 of the Public Consultation Paper makes it clear that the PDP Notice should provide sufficient information on the use of AI systems, including where the AI system makes or supports decisions affecting the data subject, generates recommendations or predictions, or provides personalised outputs or services using the data subject’s personal data.
Depending on how AI is deployed within your organisation, these circumstances may be even more familiar than they first appear, as they can easily arise in everyday business functions, including recruitment, customer engagement and marketing.
For instance, some companies deploy AI systems as part of their recruitment process. A single job posting may attract hundreds or even thousands of CVs, and reviewing and filtering these applications to determine which candidates should be interviewed can take considerable time and effort. Therefore, companies are increasingly relying on AI systems to conduct the initial screening and identify applications that should proceed for further review by HR or the hiring manager, and there is no doubt that this alone can already affect job applicants as data subjects. Depending on how the recruitment process is designed, candidates whose applications that do not meet the AI system’s screening criteria may be well excluded by the AI system before their CVs have even been reviewed by a person.
Similarly, many companies use AI systems to personalise their email marketing. For example, an online retailer may use AI to analyse a customer’s purchase history, browsing activity and product preferences to determine which products, offers or recommendations to include in that customer’s marketing emails. Instead of sending the same email marketing materials to every customer, the company uses personal data to predict what each customer may be interested in and tailor its communications accordingly. While this may make the email marketing more relevant and useful to the customer, it also involves using AI to generate predictions and personalised recommendations based on that customer’s personal data.
Therefore, where an AI system is deployed within the company, particularly where it makes or supports decisions affecting data subjects, generates recommendations or predictions, or provides personalised outputs or services using their personal data, this should be clearly explained in the PDP Notice for transparency. After all, this is also a matter of maintaining trust, and it is entirely understandable that data subjects would want to know how AI is being used to process their personal data and how that use may affect them.
3. Companies Should Plan for How Personal Data Can Be Removed from AI Systems When Consent Is Withdrawn
The third key takeaway concerns deleting personal data from AI systems when a data subject withdraws consent and requests the data controller to cease processing their personal data.
Paragraph 8.5 of the Public Consultation Paper makes it clear that, where a data subject withdraws consent and requests the data controller to cease processing their personal data, the data controller shall cease processing that personal data, including through AI systems, and take appropriate steps to delete it where it is no longer required, unless continued processing or retention is otherwise permitted under the PDPA or any other written law.
On paper, this may sound simple and straightforward that when a data subject withdraws consent and requests that processing cease, the company should stop processing and, where required, delete their personal data. However, the operational reality can be considerably way more complicated when AI is involved, because depending on how the system is designed, personal data may exist in multiple places, including the original database, training and fine-tuning datasets, uploaded documents, interaction logs, backups and databases used to retrieve information for the AI system.
Furthermore, where personal data has been used to train or fine-tune an AI model, the challenge may extend beyond locating and deleting these records. The training process adjusts the model’s internal parameters to learn patterns from the personal data, and, in some circumstances, the model may also memorise personal information. Hence, deleting the original file or removing it from the training dataset does not necessarily remove personal information retained within the trained model, and removing personal information retained within a model may therefore require more specialised measures, such as retraining the model without the relevant data or applying techniques known as “machine unlearning”, which seek to remove the influence of particular training data. One needs to understand that these measures can be technically demanding and resource-intensive, and their effectiveness will eventually depend on the model and the techniques available.
Therefore, in view of the above, companies should not be assumed that an AI system can simply be instructed to “forget” and “delete” an individual’s personal data with complete and verifiable results with ease. These challenges can become more pronounced where the AI system is maintained by a third-party service provider, because the company may be able to delete a document from its own account, while having limited to almost no visibility or control over copies retained by the AI service provider, its training datasets or the underlying AI model as a whole.
Hence, before deploying an AI system and uploading personal data, the legal team should always work with the technical team and relevant AI service providers to establish a workable process for withdrawal and deletion, and this includes understanding where the personal data will reside, whether it will be used for training, the removal mechanisms available, and the deletion support the AI service provider will subsequently provide.
4. AI Systems Should Be Appropriately Tested and Validated Before Deployment
The fourth key takeaway is that, before deploying an AI system, the company must first ensure that it has been appropriately tested and validated.
Paragraph 14.1 of the Public Consultation Paper makes it clear that, before deployment, a data controller shall ensure that an AI system that processes personal data is appropriately tested and validated to verify that it performs consistently with its intended purpose and processes personal data in accordance with the PDPA.
Paragraph 14.2 of the Public Consultation Paper further states that, where appropriate, testing and validation shall assess whether the AI system:
i. performs consistently with its intended purpose;
ii. produces outputs that are sufficiently reliable for the intended use;
iii. operates as intended under reasonably foreseeable operating conditions;
iv. presents material errors, limitations or unexpected behaviour that may affect the processing of personal data or the data subject; and
v. operates effectively with the safeguards and controls intended for deployment.
This is a straightforward yet crucial consideration as understandably, companies may be keen to deploy an AI system when its initial results appear promising, particularly where it offers significant improvements in efficiency and productivity. However, those early results should always be followed by appropriate testing and validation before the system is rolled out.
As discussed earlier, addressing problems only after personal data has already been used to train an AI model can be considerably way more complicated than correcting them beforehand.
Depending on the issue, where personal data has also been used for AI model training, remediation may require further technical measures, potentially including retraining or assessing suitable unlearning techniques. Therefore, taking the appropriate time to properly validate an AI system before rollout allows foreseeable issues to be addressed while the scope of deployment remains manageable, before those issues become embedded in day-to-day operations at scale.
5. AI Review Should Keep Pace with Changes to the AI System After Deployment
The fifth key takeaway is that companies must continue to monitor the development of their AI systems after deployment. Whenever an AI system is modified, retrained or updated, the data controller also needs to examine whether the change materially affects the processing of personal data or the intended use of the AI system.
Paragraph 18.1 of the Public Consultation Paper makes it clear that, where an AI system is modified, retrained or updated after deployment, the data controller shall determine whether the change materially affects the processing of personal data or the intended use of the AI system.
This is understandable given the pace at which AI is developing. AI systems are frequently improved, modified, retrained or updated after deployment, and their capabilities may gradually extend beyond their initial use cases.
For example, a company may initially deploy an AI tool simply to help its customer service team refine email responses, ensuring that they are grammatically accurate, clear and polite before a staff member reviews and sends them. Over time, the same system may be upgraded to access customer account information and correspondence histories, analyse enquiries, generate and send responses automatically, and even take steps to resolve certain issues, such as processing a refund or updating account details, without a staff member reviewing each action.
While these developments may offer considerable operational benefits, the processing of personal data and the intended use of the AI system may become materially different from what the company originally assessed. An AI tool introduced to assist with wording may have evolved into an AI system that accesses a broader range of personal data and independently takes actions affecting customers. Depending on the changes, the company may therefore need to revisit its earlier assessment, review the safeguards in place and, as discussed above, update its PDP Notice to accurately reflect how the AI system is being used.
This is why testing and deploying an AI system should always be followed by ongoing oversight. Practically, companies should establish periodic reviews and a process for assessing modifications, retraining or updates, so that material changes are easily identified and any necessary adjustments are made before the AI system continues to operate in its expanded usage.
6. Assess the Risks of AI Processing and Conduct a DPIA Where Appropriate
The sixth key takeaway concerns an aspect that can easily be overlooked, which is the need to conduct a Data Protection Impact Assessment (“DPIA”) where appropriate for the processing of personal data through AI systems.
Paragraphs 37.1 and 37.3 of the Public Consultation Paper make it clear that the data controller is responsible for adopting a risk-based approach to the processing of personal data through AI systems and conducting appropriate risk and impact assessments. These assessments should identify, assess and mitigate potential privacy and personal data protection risks, taking into account the nature, scope, context and purposes of the processing and the potential risks to data subjects. Where appropriate, the risk assessment shall include, or be complemented by, a DPIA, having regard to any applicable guidelines issued by the Commissioner.
It is understandable that DPIA may still be relatively unfamiliar to many companies, as Malaysia’s dedicated DPIA Guideline was only issued in early 2026, and many companies may still be working through what a DPIA involves and how to conduct one properly.
Essentially, a DPIA helps a company understand how a proposed processing activity may affect personal data protection, identify and assess the associated risks, and determine appropriate measures to address them. For example, a company may deploy an AI system to screen CVs and recommend which candidates should proceed to an initial interview. A lender may use AI to assess an applicant’s creditworthiness and affordability when deciding whether to approve a loan. Another company may deploy an AI system to manage customer service enquiries and complaints, including analysing customer records and recommending or taking action to resolve an issue.
Each of these use cases carries its own personal data protection risks, as the processing may involve profiling, generating inferences or recommendations, or making or supporting decisions that directly or indirectly affect individuals. An inaccurate or biased assessment could then affect a candidate’s employment prospects, an applicant’s access to financing or the handling of a customer’s complaint. A DPIA allows the company to examine these risks in the context of its actual use case and understand how they can be managed.
Therefore, through DPIA, depending on the findings and outcomes, companies may then implement appropriate measures include limiting the personal data processed, strengthening privacy and security controls, adjusting the AI system’s design or intended use, and introducing suitable testing, monitoring and human oversight.
7. Plan for the Retirement of the AI System and the Personal Data It Leaves Behind
The seventh and final key takeaway completes the lifecycle of personal data processing through AI, which is the retirement or decommissioning of the AI system.
Paragraphs 19.2 and 19.3 of the Public Consultation Paper make it clear that, before decommissioning of the AI system, the data controller shall determine appropriate measures for the retention, transfer, anonymisation or secure disposal of personal data. Personal data shall not be retained solely because it was used by or associated with a decommissioned AI system. Where the personal data is no longer required, it shall be securely deleted, anonymised or otherwise disposed of in accordance with the PDPA and any applicable written law.
In this article, we have discussed the use of personal data within AI systems, testing and validation before deployment, risk assessments and DPIAs, and continuous monitoring as systems are modified, retrained or updated. To complete this AI lifecycle, we must also consider what happens when the company decides to retire or decommission the system. As with any technology, an AI system may eventually be replaced or discontinued, and the company should have an appropriate exit and decommissioning plan in place.
From a company’s perspective, discontinuing an AI service may appear as simple as cancelling a subscription or closing an account. However, the underlying data protection arrangements require closer attention because the company must also determine what happens to the personal data associated with the AI system, including whether it needs to be transferred or lawfully retained, and ensure that data which is no longer required is appropriately deleted, anonymised or disposed of.
This can be particularly challenging where personal data has been used to train or fine-tune the AI model. Depending on the AI system’s design, personal data may remain in training datasets, interaction logs, backups or, in some circumstances, within the AI model itself. Therefore, removing an uploaded document or closing an account does not necessarily remove personal information retained elsewhere in the system. As discussed earlier, addressing information retained within a trained AI model may require specialised technical measures, potentially including retraining or machine unlearning.
These arrangements should be properly considered before deployment, while the company is selecting its technology and negotiating the relevant terms. As a practical matter, the company should establish what data can be returned or transferred, what must be deleted or anonymised, how retained copies and any training-related use will be addressed, and what evidence the provider can supply to confirm completion. Otherwise, simply abandoning an AI project and moving on to another AI system may leave personal data retained without an ongoing need, creating a risk of non-compliance with the PDPA.
Therefore, at the outset, companies should already agree on the relevant contractual and technical arrangements, including responsibilities for returning or transferring data, deletion procedures, applicable retention periods and confirmation that the agreed steps have been completed.
Closing Thoughts
While the Framework remains subject to finalisation following public consultation, the proposed approach already provides a useful basis for preparation. Past consultation exercises suggest that, although individual provisions may be refined, the core concepts and overall direction often carry through to the final document. Therefore, for legal, regulatory and compliance teams, this is definitely a timely opportunity to examine the Public Consultation Paper more closely, review how AI is being used across the organisation and identify any practices or safeguards that may need attention.
We will continue to follow developments closely and share further insights as the Framework progresses towards its official release.
If you have any questions on AI governance, personal data protection or the implementation of AI systems in Malaysia, please feel free to reach out to the partners in our Technology Practice Group, Ong Johnson and Lo Khai Yi, for a consultation. We have extensive experience advising organisations on technology law, data privacy and regulatory compliance matters, and would be pleased to assist your team in navigating the legal, regulatory and operational considerations arising throughout the AI lifecycle.
The Technology Practice Group of Halim Hong & Quek continues to be recognised by leading legal directories and industry benchmarks. Recent accolades include FinTech Law Firm of the Year at the ALB Malaysia Law Awards (2024 and 2025), Law Firm of the Year for Technology, Media and Telecommunications by the In-House Community, FinTech Law Firm of the Year by the Asia Business Law Journal, a Band 2 ranking for FinTech by Chambers and Partners, and a Tier 3 ranking by Legal 500.
About the authors
Ong Johnson
Partner
Head of Technology Practice Group
Fintech, Data Protection,
Technology, Media & Telecommunications (“TMT”),
IP and Competition Law
johnson.ong@hhq.com.my
◦
Lo Khai Yi
Partner
Co-Head of Technology Practice Group
Technology, Media & Telecommunications (“TMT”), Technology
Acquisition and Outsourcing, Telecommunication Licensing and
Acquisition, Cybersecurity
ky.lo@hhq.com.my.
More of our Tech articles that you should read:
- • Consumer Credit Act 2025: 10 Key Takeaways on Malaysia’s New Authorisation Regime
- • Exit and Step-In Rights in Artificial Intelligence-as-a-Service
- •Telecommunication Towers M&A: Unpacking the Transaction
