WorkThe story, in brief
Preserving the privacy of AI training data
Amazon researchers just reproduced three ways to steal private training data from AI models—and found the cryptographic fix.
The KeyNews take
Why it matters
As AI models train on increasingly sensitive data, privacy attacks and defenses are becoming a critical governance and compliance issue for any company deploying or building models at scale.
The key facts
4 to knowThree confirmed extraction attacks reproduced by Amazon Science
Cryptographic defenses identified and tested
Training data privacy governance emerging as key risk surface
Published by Amazon Science—major cloud/ML player validation
Go to the source
Amazon Scienceamazon.science
Publisher excerpt: How we reproduced three attacks that extract private training data from AI models and the cryptographic defenses that stop them.