AI News Roundup – OpenAI releases new image generation model sparking social media frenzy, Anthropic marks a court victory in AI music copyright case, North Korea boasts of AI-powered aerial drones, and more

To help you stay on top of the latest news, our AI practice group has compiled a roundup of the developments we are following.

• • This week, OpenAI released image generation features into its GPT-4o AI model, setting off a viral social media trend of AI-generated or modified images, according to the Associated Press. The model, according to the company, has the ability to accurately render text in AI-generated images, an issue that has plagued such models for years, while also adhering better to prompts. The model also can modify user-uploaded images, leading to a viral social media trend of users posting images of themselves that the model has modified to mimic the distinct style of the famed Japanese animation house Studio Ghibli. The company declined to comment. The trend also brought about discussions regarding the intersection of copyright and AI models, as they are trained on vast collections of content scraped from the internet, generally without license from copyright holders. OpenAI did not respond to inquiries regarding whether they held a license to use Studio Ghibli’s works for AI training.

• • Anthropic has won a minor victory in an ongoing lawsuit brought by several music publishers against the company on grounds of AI training-related copyright infringement, according to The Wall Street Journal. The lawsuit, brought in federal court in California by a group of record labels including Universal Music Group, alleges that Anthropic’s AI chatbot, Claude, was trained on the copyrighted lyrics of songs the labels own, and that Claude’s responses could contain verbatim reproductions of those lyrics. The labels sought a preliminary injunction that would prohibit Anthropic from training its models on those works, but U.S. District Judge Eumi Lee said that the labels failed to demonstrate that Anthropic’s conduct caused them “irreparable harm” that would justify the injunction. However, a spokesman for the labels said in a statement that “we remain very confident in our case against Anthropic more broadly,” and that the case “remains vital to protecting creators from the wholesale theft of their copyrighted works by Anthropic and other AI companies.”

• • North Korea’s state media agency KCNA recently released footage of leader Kim Jong-un supervising the deployment of AI-powered “suicide” drones to augment the state’s military capabilities, according to Reuters. KCNA also quoted Kim as stating that “the field of unmanned equipment and artificial intelligence should be top-prioritized and developed in modernizing the armed forces.” The agency also released photographs showing fixed-wing drones targeting tank-shaped targets, then exploding. AI-powered drones have become more widespread in recent years, finding use in battlefields in Ukraine, and attracting investment into military hardware startups such as Anduril Industries in the United States.

• • The Stockholm-based fashion retailer H&M announced plans to create AI-generated digital twins of thirty of the company’s models, according to BBC News. The company intends to use these twins in social media and marketing applications, but only if given permission by the models whom the twins are designed to mimic, as the models themselves would retain rights over the use of digital replicas. The company will pay the models for the use of their digital twins similarly to current arrangements that govern the use of the model’s images. Many social media platforms require the disclosure of AI-generated content, which will likely apply should H&M go ahead with their plans, but the plans still sparked concerns about possible replacement and effects on the rights of fashion workers.

• • Many of China’s AI datacenters, built out to support the country’s AI boom, stand unused, according to the MIT Technology Review. In late 2023 into 2024, the country rapidly expanded its data center capacity, fueled by both generous government support and private investors seeking to capitalize on the AI craze. Due to the massive computing costs of training AI models, the data center business model–renting out GPUs to companies for AI purposes–was more attractive. However, after China’s own DeepSeek demonstrated to the world the limitations of throwing around massive computational resources (and roiling markets in the process), the demand for such datacenters has dwindled. China’s government has still pushed AI projects to promote self-reliance and independence from foreign AI hardware, and the domestic AI industry has continued to grow, but in the meantime many of the datacenters built out will remain in limbo until either the government or other investors take charge of the facilities for other purposes.