The newer GPT-o3 and GPT-o4 mini models appear to be embedding special character watermarks in generated text. However, removing these watermarks is relatively simple, making this seem more like a short-term measure than a long-term solution