Chinese court rules AI images infringed on Ultraman copyright

If you ask an AI generator to draw a silver-and-red superhero from outer space with a glowing chest lamp, and it delivers the spitting image of Ultraman, who gets sued?

In China, the AI company takes the hit. The Guangzhou Internet Court just ruled that an artificial intelligence platform infringed on the copyright of the iconic Japanese superhero character Ultraman, marking what appears to be the world's first ruling of its kind.

According to a report by the 21st Century Business Herald, an unnamed plaintiff holding partial rights to Ultraman in China took an undisclosed AI firm to court after its software generated images bearing striking resemblances to the character.

The court determined that the generated outputs were substantially similar to Ultraman, which suggested that the protected character had been fed into the model during training. The court ordered the AI firm to pay 10,000 yuan, roughly $1,400, in damages.

A tale of two courts

This decision throws a spotlight on a glaring contradiction inside China's fast-moving legal framework for AI.

Just last November, the Beijing Internet Court handed down a landmark ruling granting copyright protection to an AI-generated image. In that case, the judge ruled that generative AI was merely a tool helping a human artist create.

Yet Guangzhou took the opposite stance, finding that the AI actively participated in creating the content rather than serving as a passive paintbrush. As legal blogger Beijing Lawyer Zhao Duidui noted, the two verdicts clash head-on, proving courts are still struggling to define how generative models actually function.

Zhou Chengxiong, a director at the Chinese Academy of Sciences, warned via the Global Times that rulings like Guangzhou's could scare Chinese tech firms away from investing in foundation models out of legal fear.

Why it matters

We think this ruling sets a massive precedent because the court bypassed the technical black box entirely. Instead of requiring proof that scraped files lived inside training logs, the judge inferred infringement simply because the generated output looked substantially similar to the original character.

For generative AI builders, this means prompt safeguards are no longer optional. If your image model can cough up copyrighted characters on demand, you are on the hook, regardless of whether you claim the system is just an open-ended engine.

For IP holders, getting a court to recognize algorithmic infringement is a psychological win, even if a $1,400 penalty barely covers coffee money.

As Forbes Council Member Lindsey Witmer Collins argued regarding the New York Times lawsuit against OpenAI, the generative AI boom suffers from a fatal lack of traceability. Until AI platforms build transparent data tracing where original creators are credited and compensated, courts worldwide will continue to issue conflicting, ad-hoc rulings.