When AI-generated images in the distinctive hand-drawn style of Studio Ghibli flooded social media earlier this year, it exposed a fundamental tension in modern creativity: the ease with which generative tools can now replicate identifiable artistic styles, and the near-impossibility of legal protection when they do. Sony is proposing one answer.
According to reporting by The Nikkei, Sony AI, the technology conglomerate's research division, has developed what it calls Protective AI. The system works by training on copyrighted material—in this case, Studio Ghibli films—so it learns precisely what to refuse to generate. The goal is to block AI systems from producing unauthorised imitations of protected work, even when users try indirect prompts to circumvent restrictions.
This approach reveals both the ambition and the limits of technical solutions to copyright problems. Rather than stopping AI from being trained on copyrighted material in the first place, Protective AI accepts that as a given and builds guardrails downstream. The second component aims to create compensation mechanisms; creators and rights holders would receive payments when their work contributes to AI-generated outputs.
The appeal is obvious. Sony Group controls an enormous portfolio spanning games, music, films, and anime partnerships. A tool that could protect those assets while establishing a revenue stream for creators represents genuine financial incentive for the company. For artists and studios watching their styles replicated at scale, the prospect of at least being paid would be an improvement on the current situation where AI companies invoke fair use while generating content that competes directly with human creators.
But several hard problems remain unresolved. First, Sony's previous research on protecting music creators' rights suggests technical tracking is possible but complex. Determining how much compensation flows to whom when thousands of training images contribute to a single output is not a solved problem. Second, Protective AI is voluntary. Unless competing platforms adopt similar protections or regulation mandates them, studios and artists who don't use Sony's system remain vulnerable. An artist's style is not currently protected by copyright law in most jurisdictions; the works themselves are, but replicating a recognisable aesthetic falls into a legal grey zone.
Third, the tool addresses symptoms rather than causes. The core issue is that AI models are trained on copyrighted material without creator consent or compensation. Protective AI doesn't prevent training on such material; it merely adds a refusal layer. As long as companies prioritise speed and scale over rights clearance, new models will continue to absorb protected content. Some argue that intellectual property law needs reform to require licensing or compensation at the training stage, rather than hoping individual tools will police the outputs.
That said, Sony's investment suggests the industry is beginning to accept what many creators have long insisted: that using people's work to build profitable products should come with accountability. The Anthropic copyright settlement in late 2025, in which the AI company agreed to pay authors compensation for using their works in training, marked a shift toward recognising that principle, even if the legal mechanisms remain contested.
For Australian creators, the implications are mixed. International copyright frameworks like the Berne Convention protect Australian artists' works abroad, but enforcement remains patchy. Tools like Protective AI could offer some protection if widely adopted. More likely, the real shift will come through litigation and regulation, not through technical Band-Aids applied by individual companies.
Sony's announcement reflects genuine corporate responsibility, but it also reflects something simpler: that in the absence of clear legal rules, large companies are building their own. Whether these proprietary systems are enough, or whether they'll merely create a patchwork where only companies with resources can protect their interests, remains to be seen.