<p>The present research proposes a high-capacity reversible data embedding technique (RDH-EI) for applications utilizing cloud storage that use a recursive look ahead adaptive MSB prediction approach. The paper also proposes a three-level image encryption approach where on the first level, the image is subdivided into 4 × 4 blocks and the blocks are permuted. In the second level, two different neighbourhoods namely the middle neighbourhood and peripheral neighbourhood are estimated, from which the elements of the middle neighbourhood are permuted within the respective neighbourhood. In the third level, the elements of the peripheral neighbourhood without including the elements of the middle neighbourhood are permuted within the respective neighbourhood. The image that was encrypted by the image owner was uploaded to the cloud. Out of five different neighbourhoods estimated on the 4 × 4 blocks, the best neighbourhood is then estimated from the encrypted image which provides a maximum embedding capacity. The data is then embedded in the best neighbourhood using the adaptive MSB prediction algorithm. Once the embedding iteration is completed, the marked encrypted image obtained on the previous iteration is again evaluated for the best neighbourhood. This recursive process is repeated till the embedding capacity of the best neighbourhood is less than the threshold. The standard test images from the datasets BOSS base, UCID and BOWS-2 were used to evaluate the proposed approach using measures like embedding rate, SSIM and PSNR. The suggested method offers an average embedding rate that is greater than previous recent RDH-EI approaches, at 3.714 bpp, 3.4826 bpp and 2.9412 bpp for the datasets BOSS base, BOWS-2 and UCID, respectively.</p>