It is impossible to tell if the user owns the copyrighted video without extensive identity checks which would be impractical. Does openai assume the input videos to be owned by the requestee and as such I should not worry about my API key ever getting blocked for copyright reasons? My assumption is it is the user’s responsibility to check if he is allowed to use content within the legal system he exists in, and just checking this wouldn’t cause any red-flags to my account if user’s started inputting a lot of youtube videos to the whisper API, for example.
For your own interpretation:
They don’t block “your API key”, they ban your account. This has mostly been seen when jailbreak prompts from known violating sites were used, or repeated violating content was not sent to moderations for pre-detection or was stimulated out of the AI.
For Whisper, you don’t have the option of not receiving back offensive violent suicidal threatenting content transcribed - the AI has to make the censoring decision, if any.
It would be good to run moderations on that output and shut down the user satisfaction in receiving it back, though.
Copyright infringement, liability, etc depends on what you or users do with the derived output and the dependency on other creative work. There is no “copyrighted content fingerprint detection” so far.