21 results found
Privacy-preserving AI models that process encrypted data may be vulnerable to undetectable manipulation, researchers warn. The finding challenges assumptions about security in confidential computing.