What sensitive personal data might AI models expose?

Question

Accepted Answer

Sensitive Personal Data Potentially Exposed by AI Models

AI models could potentially expose sensitive personal data including names, email addresses, transaction and location records, as well as video, voice, and other recordings. Organisations in Singapore developing AI technologies, particularly those using personal data for training generative AI models, may soon be required to provide clear, AI-specific notices to affected users under proposed guidelines. These notices are intended to address concerns about personal data being exposed, reconstructed, or disclosed to third parties through generative AI.
Types of Personal Data at Risk

The types of personal data that could be exposed by generative AI models are varied and include identifying information such as names and email addresses. Beyond basic contact details, more sensitive data categories like transaction records and location history may also be compromised. Furthermore, the potential for AI models to expose video, voice, and other forms of recordings raises significant privacy concerns for individuals.
AI-Specific Notices and User Information

Under proposed guidelines from Singapore’s Personal Data Protection Commission (PDPC), organisations are being encouraged to issue AI-specific notices. These notices should detail what an AI model does, specify the types of personal data it utilises, and outline the procedures for users to opt out or withdraw their consent. For instance, a company developing a text-to-speech service is expected to inform users that their voice recordings will be used for model training and explain how this data aids in speech pattern recognition.
Methods of Notification and Consent Requirements

The proposed methods for delivering these AI-specific notices include in-app pop-ups, dedicated webpages, updates to privacy policies, or emails with opt-out links. However, it remains unclear whether these guidelines will mandate explicit user consent or if providing an opt-out option will suffice. The applicability of these requirements when personal data is anonymised before AI model training is also yet to be specified.

Reading history

Hide →

This history is cleared when the page is closed.

What sensitive personal data might AI models expose?

Loading

Was this article helpful?

Helpful

Not helpful

Report an issue

1

0

0

Share with friends