The only parts of the article that seem to be news are a) OpenAI's blanket consent is very broad and doesn't warn users what might be done with cameos, or segment consent into various different types of content use (as even 25yo modeling sites do) b) that subset of users will bypass the guardrails and c) OpenAI doesn't close the feedback loop by notifying the users in a) what the users in set b) are doing, let alone allow revising or revoking consent.
But why is the conversation only about b) (the predictable bad behavior by users) rather than a) and c) (feasible solutions)?
Notopoulos correctly remarks: "part of an overall pattern of how OpenAI has approached the concept of copyright and intellectual property: asking forgiveness, not permission."
By the way previous (non-sexual-content) incarnations of this sort of issue are 2019 when Clearview scraped 3 billion images non-consensually from people's social-media and state DMVs [https://news.ycombinator.com/item?id=35421117].
Or the previous 2024 OpenAI/Scarlett Johansson-sounding voice shenanigans. Or the existing proliferation of AI porn elsewhere.
OpenAI is still doing basic experiments with which product offering are well received by users and/or work well and which are not. If some data provided by users (e.g. photos depicting the user) are seen to be very essential to the success of the AI-created content using this data so that OpenAI will likely loose an insane amount of money of these users leave (I think this is rather unlikely, but not impossible), then OpenAI will think about some concept of revenue-sharing, but not before (why should they?).
> Is the author suggesting we should stop people from creating fetish content of purely AI-generated characters?
Presumably not, but she's farming outrage rather than suggesting any fix. In the above suggested setup, people could then generate fetish consent from the much smaller set of users who consented to have fetish consent generated from them. But then of course they might expect some royalties or revenue-sharing, or at least identification/attribution/watermarking so the depicted user could drive traffic to social-media. OpenAI is skirting around not just segmented consent but any concept of revenue-sharing (i.e. OpenAI wants to dip its toe into OnlyFans territory, but without any revenue-sharing or licensing deal with creators).