AI agents with access to decrypted messages pose new challenges for end-to-end encryption, raising questions about endpoint privacy, operating-system permissions and whether users can protect conversations when their correspondents enable AI assistants.