Attribute-based encryption (ABE) allows one-to-many encryption with static access control. In many occasions, the access control policy must be updated and the original encryptor might be required to reencrypt the message, which is impractical, since the encryptor might be unavailable. Unfortunately, to date the work in ABE does not consider this issue yet, and hence this hinders the adoption of ABE in practice. In this work, we consider how to efficiently update access policies in Ciphertext-policy Attribute-based Encryption (CP-ABE) systems without re-encryption. We introduce a new notion of CP-ABE supporting access policy update that captures the functionalities of attribute addition and revocation to access policies. We formalize the security requirements for this notion, and subsequently construct two provably secure CP-ABE schemes supporting AND-gate access policy with constant-size ciphertext for user decryption. The security of our schemes are proved under the Augmented Multi-sequences of Exponents Decisional Diffie-Hellman assumption. Abstract. Attribute-based encryption (ABE) allows one-to-many encryption with static access control. In many occasions, the access control policy must be updated and the original encryptor might be required to re-encrypt the message, which is impractical, since the encryptor might be unavailable. Unfortunately, to date the work in ABE does not consider this issue yet, and hence this hinders the adoption of ABE in practice. In this work, we consider how to efficiently update access policies in Ciphertext-policy Attribute-based Encryption (CP-ABE) systems without re-encryption. We introduce a new notion of CP-ABE supporting access policy update that captures the functionalities of attribute addition and revocation to access policies. We formalize the security requirements for this notion, and subsequently construct two provably secure CP-ABE schemes supporting AND-gate access policy with constant-size ciphertext for user decryption. The security of our schemes are proved under the Augmented Multi-sequences of Exponents Decisional Diffie-Hellman assumption.