AI watermarks will always be trivial to archival data

If everyone is using ai (per reported revenues) and nobody like the outcome (banned here, dismissed there), then what is the DCT swizzling functionality designed for in the hardware originally? The multimodal abilities are great, but if you deal with text only, what is the future of AI? If everyone is using ai (per reported revenues) and nobody like the outcome (banned here, dismissed there), then what is the Pareto frontier of loss vs complexity?