Putting consent at the centre of the test means a release form becomes a budget line, while an unbounded personality claim stays a legal exposure [2]. Notice-triggered exposure for the platforms that generate the content [6] is the familiar shape of notice-and-takedown, and it tells an operator what compliance costs to build: a consent record that survives an evidentiary challenge, plus a queue that clears complaints faster than claimants can file them. Because a model's own fabrications are covered too [7], that queue has to cover output nobody asked for.
Then the price. The Beijing Internet Court's first deepfake ruling, in June 2024, fined an app operator 3,500 yuan, about $482, for letting users run face-swap templates on two models without permission [10], an implied 7.26 yuan to the dollar [1] and a figure closer to a filing fee than a deterrent. Push it through one platform group's own housekeeping: the 85,000-plus videos ByteDance says it has pulled since January for unauthorised reproductions of faces and voices [13] would carry 297.5mn yuan, roughly $41mn, at that rate [2], across one company and eight months.
Private litigation cannot carry the volume. The Guangzhou Internet Court has handled about 700 face-theft cases in three years, according to National Business Daily [11], which is some 233 a year, one every 1.6 days [3]; more than 95% of the 128,000 microdramas released in the first quarter of 2026 used AI somewhere in production [14], meaning at least 121,600 titles, about 1,350 a day [4]. The channel that scales is administrative, and the Cyberspace Administration's September 2 tally of 5.61mn items and roughly 49,000 accounts across 2,400 sites and apps [12] works out at about 114 removals per account [5].
The unfair-pricing half of the guidance has the cleaner evidentiary path, because the pricing log is the exhibit. "We cannot expect every consumer to become an expert at spotting deception," Zhou Jiahai, who heads the court's research office, told Xinhua [9].
So the consent line is worth building against, because the compliance artefact is cheap relative to the exposure and because a supply side is already forming, with Joey Wong, Lawrence Ng and ordinary people licensing their likeness to AI platforms [15]. The counter-thesis sits in the same arithmetic: at 3,500 yuan a ruling and 233 rulings a year against 1,350 AI-assisted titles a day, skipping consent stays cheaper than buying it for anyone whose subject will not sue, and the market becomes a market for famous faces only. What would settle it is the size of the next awards, since the guidance replaces case-by-case discretion [17] without, on this account, setting a damages scale.
