International Association
for Cryptologic Research

Homomorphic encryption (HE) enables computation delegation to untrusted third-party while maintaining data confidentiality. Hybrid encryption (a.k.a Transciphering) allows a reduction in the number of ciphertexts and storage size, which makes HE solutions practical for a variety of modern applications. Still, modern transciphering has two main drawbacks: 1) lack of standardization or bad performance of symmetric decryption under FHE; 2) lack of input data integrity. In this paper, we discuss the concept of Authenticated Transciphering (AT), which like Authenticated Encryption (AE) provides some integrity guarantees for the transciphered data. For that, we report on the first implementations of AES-GCM decryption and Ascon decryption under CKKS. Moreover, we report and demonstrate the first end-to-end process that uses transciphering for real-world applications i.e., running deep neural network inference (ResNet50 over ImageNet) under encryption.