Singapore’s Infocomm Media Development Authority (IMDA) has introduced voluntary guidelines for generative artificial intelligence (GenAI) chatbots, encouraging companies to implement “nutrition labels” to improve transparency. These infocards, which can take the form of disclosure documents or dedicated web pages, aim to provide users with clear, consolidated information regarding the capabilities, data privacy practices, and safety limitations of AI applications.
Standardizing Transparency Through AI Infocards
The IMDA framework treats AI transparency much like food labeling, according to Minister for Digital Development and Information Josephine Teo. By providing a single, accessible location for essential facts, the guidelines aim to simplify complex terms of service and privacy notices that are often scattered across various platforms.
According to the IMDA, these infocards should address core user concerns, including:
- What the chatbot can and cannot perform.
- Safety and reliability metrics.
- Protocols for personal data handling.
- Methods for reporting issues or raising concerns.
While the framework is currently voluntary, the public sector is set to lead adoption. Agencies such as the National Library Board and the Health Promotion Board have announced plans to implement these disclosures. Major private organizations, including DBS, Google, Meta, OCBC, and Singapore Airlines, have also signaled their intent to align their transparency practices with these guidelines over the next six to 12 months.
Clarifying Data Collection for Model Training
Alongside the infocard guidelines, the IMDA released updated advisory notes regarding the Personal Data Protection Act (PDPA) as it applies to GenAI development. The guidance specifically addresses the common practice of using customer data—such as call recordings—to train AI models.
The IMDA clarifies that general statements regarding “product development” or “personalization of services” are no longer sufficient for regulatory compliance. Organizations must now explicitly notify individuals if their personal data is being used for GenAI training. Furthermore, companies must provide clear mechanisms for individuals to decline or withdraw consent.
Navigating Web Scraping and Public Data
The new advisory clarifies the boundaries of web scraping. Organizations are permitted to scrape publicly accessible personal data without consent for the purpose of developing GenAI models. However, the IMDA warns that this does not apply to data protected by digital barriers.
If data is hidden behind paywalls, registration requirements, or other access controls, it may not qualify as “publicly available.” In these instances, organizations are required to obtain explicit consent. The authority also notes that individuals retain the right to request access to or correction of their personal data, even after that data has been incorporated into a GenAI model, though the IMDA acknowledges the technical difficulty of such requests given the scale of modern datasets.
Did you know? The IMDA suggests that companies adopt upstream data handling and case-by-case review processes to manage user requests for data correction in AI models effectively.
Frequently Asked Questions
Are these AI nutrition labels mandatory?
No, the current IMDA guidelines are voluntary. However, several government agencies and major corporations have committed to adopting them within the next year.
Can companies use my data for AI training without asking?
Under the new guidelines, organizations must be specific. Broad phrases like “service improvement” are insufficient. Companies must explicitly state if data is used for GenAI development and explain how users can opt out.
What happens if I want my data removed from an AI model?
The guidelines affirm that individuals have the right to request access to or correction of their personal data. The IMDA encourages companies to implement technical measures and tracking to facilitate these requests, even within complex AI training environments.
Have you noticed changes in the privacy notices of the AI tools you use daily? Share your thoughts in the comments below or subscribe to our newsletter for the latest updates on AI regulation and digital policy.