How AI Training Could Compromise Personal Privacy
A new report says that AI models trained on personal data can sometimes reveal details they memorised, even when the data was labelled anonymised. Researchers say such details could help someone identify individuals and build profiles of them. The piece raises a concern about how well anonymous data stays private once an AI model has learned from it.
Context
Concerns about AI and privacy often centre on whether personal data used in training can be pulled back out later by clever questioning. Anonymising records is one common safeguard, but it may not be a complete shield.
Burning Potato score: 81%