Tracing the source finally produced an invoice
On July 19 I wrote about what Eggers did and didn't say when he accused OpenAI of stealing voices. Today's court approval in the Anthropic case shows, for one instance of that same problem, exactly where the exhaust went: books downloaded from pirate sources like Library Genesis and Pirate Library Mirror are now being repaid at $3,000 per work.[1], [3]
But I want to underline this: Judge Alsup's finding that training counts as fair use binds only this case. The suits against Google, Meta, Midjourney and OpenAI are still open — so this $1.5 billion closed one end of a pipe, not the pipe itself.[1]
YouTube's 'inauthentic content' definition is also a tracing problem
The same day, YouTube defined three categories of content that can't be monetized; the third one caught my attention most — content using AI personas to discuss sensitive topics like health, finance or law. Trust and safety chief Matt Halprin's line lands exactly where I look: 'That exact same tool can allow you to make lots of... very generic... content farming.'[2]
If it holds up, this policy doesn't tell the viewer who or what is actually speaking; it only tells the advertiser 'this can't earn money.' It doesn't label the source itself, it just cuts the payout — which is a revenue-protection measure, not transparency.[2]
Next question: who consented, who got paid?
My skepticism is an occupational deformation, I know. But putting the two stories side by side, here's what I see: on one side a court ruling that pays money back, on the other a platform policy that withholds money — both still solve the source-tracing problem in the currency of money, not the currency of consent.[1], [2]
What I'll test next: when one of the Google, Meta, Midjourney or OpenAI suits resolves, will the payout again be a flat per-work number, or will it actually produce an answer to who, when, with what consent? We may still never fully learn where the exhaust goes.[1]