Taiwanese YouTuber Tsai A-ga (蔡阿嘎) triggered the controversy by placing a painting a Japanese street artist had made for his children side by side with an AI-generated image and inviting followers to vote on "who wins, AI or the painter"CITE:E1. Japanese media outlets picked up the story afterward and described the comparison as "too rude" to the human artistCITE:E2.
As the dispute widened, an online boycott campaign gathered momentum, producing figures large enough to draw brand partners into the fallout:
| Metric | Figure |
|---|---|
| Boycott petition signatures | Over 100,000 |
| Users who publicly responded | More than 120,000 |
| Partner brands implicated | More than 20 |
The petition surpassed 100,000 signatures, drew more than 120,000 public responses, and implicated more than 20 of Tsai's partner brandsCITE:E3.
The US Copyright Office and Taiwan's Intellectual Property Office reach a similar conclusion: protection turns on human creative input, not on AI output by itself. The US Copyright Office holds that content generated purely by AI, with no identifiable human creative expression involved, is generally not eligible for copyright protectionCITE:E4. Taiwan's Ministry of Economic Affairs Intellectual Property Office takes a comparable position: output that an AI system generates independently, where a human only issues a simple instruction, is not protected by copyright, while work involving human creative choice and arrangement may still qualify for protectionCITE:E5.
AI image generation carries risk at two distinct points: when copyrighted works are fed into a model during training, and when a generated output resembles an existing copyrighted work. Feeding copyright-protected works into an AI system for training constitutes reproduction under copyright law and, in principle, requires authorization from the rights holderCITE:E6. On the use side, a case from the Guangzhou Internet Court in China shows that when an AI-generated image is substantially similar to an existing copyrighted character, the user who generated it may bear infringement liabilityCITE:E7.
Two protective mechanisms are now emerging: an opt-out right for training data and opt-in licensing for commercial use. The European Union's AI Act grants rights holders the right to opt out of having their works used for AI trainingCITE:E8. Separately, an opt-in commercial licensing model has appeared in the music industry: Universal Music Group (UMG) partnered with AI music company Udio, with rights holders actively licensing their works for useCITE:E9.
The Tsai A-ga backlash and the legal record point to the same gap from opposite endsCITE:E1CITE:E4. Authorities in the US and Taiwan withhold copyright from AI output that lacks human creative inputCITE:E4CITE:E5, while treating the training data that makes such output possible as a reproduction that in principle requires authorizationCITE:E6. That asymmetry — AI training draws on human works, but AI output rarely earns human-style protection, and can itself create infringement exposure for the userCITE:E7 — is what opt-out mechanisms like the EU AI ActCITE:E8 and opt-in deals like UMG's arrangement with UdioCITE:E9 are now trying to close, each from a different direction: one gives creators a right to refuse, the other gives them a right to be paid.
The Tsai A-ga episode and the copyright rulings point to the same structural problem from two ends. On one end, the US and Taiwan IP authorities draw copyright protection at human creative input, leaving pure AI output outside the law entirely — which means a "who wins" poll comparing a protected human painting to an unprotected AI image was already legally lopsided before the ethics debate started. On the other end, training itself is being treated as reproduction requiring authorization, which is why opt-out and opt-in mechanisms are being built at the input side rather than the output side. The metric worth tracking next: whether more industries follow the opt-in licensing path UMG took with Udio, or whether opt-out regimes like the EU AI Act's stay the default — because that choice determines whether creators get paid upfront or must actively police infringement after the fact.
South Korea's Ministry of Oceans and Fisheries has picked Ulsan as the pilot site for an underwater data center designed to scale toward 100,000 servers, backed by a 51.1 billion won (about $35.4 million) budget, with testing targeted for this year and commercial-zone construction set for 2031.
NVIDIA has restructured its financing backing for OpenAI's Ohio data center campus, with the originally discussed $250 billion guarantee narrowed to below $120 billion for the first phase and finalized at $105 billion for a 20-year lease. The revised deal covers roughly 5GW of an eventual 8GW-to-10GW site; funding for the remaining 5GW is still undecided.
Anthropic's annualized revenue run rate reached $65 billion by the end of July 2026, roughly seven times higher than a year earlier and above rival OpenAI's most recently reported $40 billion run rate<CITE:E1><CITE:E6>. The company shared the figures with investors over a weekend operational update, alongside a $965 billion valuation following its May Series H round and a June SEC filing for an IPO<CITE:E2><CITE:E16><CITE:E3>.