AIME: Watermarking AI Models by Leveraging Errors | IEEE Conference Publication | IEEE Xplore