Informally, Privacy Homomorphism (PH) refers to encryption schemes with a homomorphic property allowing to obtain Ek(a + b) or E_k(a ? b) from ciphertexts E_k(a) and E_k(b) without the knowledge of the decryption key. PH has a wide range of applications in information security due to its homomorphic property, but the best achievable security of them is yet unknown. In this paper, we discuss the security of PH in the black-box model and find that any PH is at most semantically secure under non-adaptive chosen-ciphertext attacks (i.e. IND-CCA1 secure). We also show that the IND-CCA1 security can be achieved with a small amount of hardware, namely, we offer a hardware-based solution to construct PH with provable security.