Advisory Guidelines on Use of Personal Data in Generative AI
Singapore · Guideline · Personal Data Protection Commission
High impact. A reviewer confirms each change before it is listed here.
Our summary
These guidelines clarify how Singapore's Personal Data Protection Act applies to the development, deployment and post-deployment of generative AI systems. They address the collection and use of personal data for training generative AI models, including when organisations can rely on exceptions such as the publicly available data exception, and set out data protection responsibilities across different stakeholders in the generative AI lifecycle.
Written by the Index in our own words. It is not legal advice; the official text is the authority.
Inside this policy
Every section of this policy in order. Our notes summarise each one in plain language and do not quote the official text, which is the authority.
Drafted by the Index An automated summary of the official text, not yet reviewed by a person.Reviewed Checked by a person against the official text.
4 sections
Jump to a chapter
Part I Introduction and scope
This advisory and non-binding guideline explains how the PDPA applies to generative AI systems, clarifying when organisations may use personal data without consent for AI system development, testing and monitoring, how to be transparent about AI system data use, and how third-party providers may support implementation. The guideline addresses key data protection issues including collecting and using personal data to develop models, allocating data protection responsibilities across the AI lifecycle, and handling individual requests about personal data processing, organised by stages: development (publicly available data, notification and consent, protection), deployment (accountability, retention, protection, purpose), and post-deployment (access and correction requests).
Part II Development, collecting and using personal data to
Organisations can collect personal data without consent under the Publicly Available Exception if data is generally accessible with few or no restrictions, including web-scraped data, subject to a reasonableness test considering digital barriers' purpose, effect, complexity, and data accessibility elsewhere; examples of publicly available data despite digital barriers include government registers available for a fee, paywall-monetised news websites, and forums requiring simple registration, with organisations assessing borderline cases required to document reasoning in a Data Protection Impact Assessment or written record and be prepared to justify to the Commission. For data provided by individuals or created through their use of the organisation's services, organisations must obtain consent unless relying on exceptions, and must notify individuals of collection and use purposes before collecting data and of new purposes before using data for them, with significant departures from past commercial activities potentially requiring fresh consent. Organisations developing large-scale generative AI model training or fine-tuning must provide explicit notifications specifically mentioning generative AI model training rather than generic statements, describing model functions, affected personal data categories, how data will be used in training, and mechanisms for declining or withdrawing consent, while layering information or combining notifications for multiple models is permitted; organisations must not require individuals to consent to uses beyond what is necessary to provide a product or service as a condition of receiving that product or service, and are encouraged to anonymise datasets and minimise data collection, implementing appropriate controls to protect personal data where it remains necessary.
Part III Deployment - data protection responsibilities of generative
Model Providers, System Providers, and System Deployers must comply with PDPA obligations in their respective roles: Model Providers must stop retaining personal data or remove means of association when the original collection purpose no longer applies, though may retain training data longer if they document the rationale and regularly review necessity, and must protect data from downstream users through reasonable security arrangements and document security measures, access controls, residency and retention policies, and incident procedures; System Providers must periodically review and implement additional security arrangements to address risks like prompt injection attacks and share information on safeguards with downstream deployers. System Deployers bear primary responsibility for ensuring systems meet PDPA obligations, must assess upstream safeguards during procurement, define intended processing purpose and amount of personal data required, configure access controls aligned with purpose, safeguard new data sources like prompts and outputs, educate end users on appropriate data inputs, develop and document clear written safeguard policies, and regularly review safeguard sufficiency especially for systems with agentic functionalities, while carefully considering and being transparent about privacy-utility trade-offs by implementing role-based data controls, data classification systems and escalation protocols for complex tasks.
Part IV Post deployment - addressing individual requests about
Organisations must respond to individual requests for access to and correction of personal data collected, used or disclosed for generative AI development or deployment unless a PDPA exception applies. Best practices include verifying data accuracy at collection, implementing data cleaning techniques and maintaining data provenance records; reviewing requests case-by-case and removing inaccurate data before future training; and tracking and progressively adopting technical measures to remove inaccurate data from models, using output filters in the interim.
Controls
What this policy requires, forbids, grants or penalises, one control at a time and in our own words. Each control links to the official text, which is the authority. This is not legal advice.
Drafted by the Index Drafted automatically from the official text, not yet reviewed by a person.Reviewed Checked by a person against the official text.
Themes covered 0 of 35
Choose a theme to see only its controls. About the themes
Risk and scope 0 of 4
- Risk classification, not covered
- Prohibited practices, not covered
- Risk management system, not covered
- Impact assessment, not covered
Data and privacy 0 of 4
- Data governance and quality, not covered
- Data subject rights and consent, not covered
- Cross-border data transfer, not covered
- Copyright and training data, not covered
Transparency and records 0 of 4
- Transparency and disclosure to users, not covered
- Labelling of synthetic content, not covered
- Technical documentation, not covered
- Record keeping and logging, not covered
Technical safeguards 0 of 4
- Human oversight, not covered
- Accuracy, robustness and cybersecurity, not covered
- Bias and non-discrimination, not covered
- Testing and evaluation, not covered
Assurance and lifecycle 0 of 4
- Conformity assessment and certification, not covered
- Registration and notification, not covered
- Incident and breach reporting, not covered
- Post-market monitoring and corrective action, not covered
Rights and protected groups 0 of 4
- Automated decision-making rights, not covered
- Children and vulnerable groups, not covered
- Biometric identification and surveillance, not covered
- Whistleblowing, complaints and redress, not covered
Roles and organisation 0 of 6
- Supply chain and provider-deployer duties, not covered
- General purpose and foundation models, not covered
- Accountability and governance roles, not covered
- Training and AI literacy, not covered
- Public sector use and procurement, not covered
- Innovation support and sandboxes, not covered
Enforcement and sectors 0 of 5
- Enforcement, penalties and regulators, not covered
- Sector rules: health, not covered
- Sector rules: finance and insurance, not covered
- Sector rules: employment, not covered
- Sector rules: education, not covered
The full list of controls could not be loaded, so only the first 0 are shown.
Change history
No changes recorded since we began monitoring this policy.