[mattbee]

If you had the prompt and the output, and you knew your work was in the training data, you _could_ retrain the whole model minus your work, and then feed the same prompt back in.

Then if the output looked nothing like your work as it did for the inclusive model, you could prove your work was influential, possibly to the point of a © claim.

Probably quite an expensive way to do it, unless it's somehow possible to reverse specific inputs from a trained model :)