Closed-source AI models often incorporate safety filters that can impede work on security software by blocking certain queries or tasks