Researchers develop method to erase sensitive data from AI models

Researchers have developed a method allowing AI models to erase sensitive data without access to the original training set. By using surrogate data and calibrated noise, the technique effectively removes unwanted information while preserving model utility. This approach significantly reduces the computational costs and energy intensity typically associated with retraining, offering a more efficient solution for data deletion. This innovation directly addresses growing privacy concerns and regulatory demands under laws like GDPR and CCPA. It provides a provable way for individuals and organizations to ensure personal or copyrighted material is removed from AI systems. The method maintains strong privacy protections comparable to full retraining, making it practical for real-world applications where source data is unavailable or difficult to recover. The technology is highly relevant to open data discussions as it balances transparency with privacy rights. It enables the responsible use of AI by ensuring that sensitive information does not persist in open models indefinitely. This advancement supports ethical AI practices, allowing for the continued development and sharing of open models while respecting individual privacy and legal compliance requirements.

Source: thedailystar.net
Published on 2025-09-01

Researchers develop method to erase sensitive data from AI models

Related news