I would not be surprised if this already happens on the OpenAI back end but the attack surface is immense and false positives will damage the platform quality,

I would not be surprised if this already happens on the OpenAI back end but the attack surface is immense and false positives will damage the platform quality, so it will be hard to solve 100% given we have no concept of how many ways it can be done.