The Hal Times

Sony Music and Warner Chappell have filed a lawsuit against Anthropic, accusing the company of a broad campaign of intellectual property theft. The suit targets the alleged unauthorised use of copyrighted works in training large language models.

The plaintiffs demand statutory damages of up to $150,000 for each infringed work. The case centres on claims that Anthropic’s models were built using pirated or improperly licensed material, a charge the company has not yet publicly addressed.

The action follows a pattern of legal challenges from content owners against AI developers. It remains unclear how courts will interpret fair use in the context of model training, or whether the damages sought will be sustained.

The suit underscores the unresolved tension between the data needs of large language models and the rights of copyright holders. (Inshorts; TechCrunch)

Anthropic Opens Model Hardware Standard Research Preview

Anthropic has released a research preview of the Model Hardware Standard, a shared specification intended to let AI agents safely operate physical devices. The standard aims to establish common interfaces between models and hardware, reducing fragmentation in robotics and embedded systems.

The move follows growing interest in deploying large language models beyond text generation. Whether the standard gains industry adoption remains to be seen, as competing proposals already exist.

Anthropic Study Finds Models Improve At Training Other Models

A study by Anthropic reports that AI models are becoming more effective at training other models. The research suggests iterative self-improvement may reduce reliance on human-curated datasets, though the long-term stability of such systems is untested.

The findings add to a body of work on recursive model refinement. Whether the gains translate to practical deployment at scale is not yet established. (The Indian Express)

From the Laboratories

•  Neocloud Lambda secures $1B in debt to purchase Nvidia AI chips for leasing to Microsoft. (TechCrunch)

•  Nvidia’s AI advantage is shifting toward data centre traffic control rather than raw GPU power. (TechCrunch)

Copyright Disputes

Music Publishers Sue Anthropic Over Training Data

A collective of major music companies, including Sony Music and Warner Music, has filed a lawsuit against the AI firm Anthropic. The plaintiffs allege that the company used protected song lyrics from artists such as Mariah Carey and Taylor Swift to train its large language models without permission or compensation.

The legal action focuses on the reproduction of copyrighted material during the training process. The plaintiffs contend that Anthropic infringed upon their intellectual property rights by ingesting these lyrics into its systems. They seek to address how AI companies handle creative works.

The outcome of this litigation remains uncertain, though it marks a significant attempt by the music industry to establish legal boundaries regarding AI training practices. Anthropic has not yet issued a detailed response to the specific claims presented in the filing. (news.sbs.co.kr; 헤럴드경제)

South Korea Appoints Leaders For National AI Strategy

Lee Jae-myung has appointed Lee Hae-min and Ha Jung-woo to oversee South Korea's artificial intelligence strategy. Ha Jung-woo, who returns to the national AI strategy committee, is tasked with coordinating the country's AI G3 plan to bolster its competitive standing.

The appointments reflect a push to solidify national policy in the sector. The committee aims to streamline governmental efforts as South Korea seeks to influence global AI development standards and industrial applications. (CHOSUNBIZ; 디지털투데이)

Arizona Supreme Court Denies Proposed AI Restrictions

The Arizona Supreme Court has rejected a proposal that sought to ban the use of AI tools by judges. The move comes as state courts navigate the integration of new technologies within the legal system.

The proposal aimed to limit the influence of automated systems in judicial decision-making. By rejecting the ban, the court maintains current flexibility regarding the tools available to the bench, leaving the long-term governance of AI in Arizona courts an evolving matter. (Havasu News)

In Brief

•  A South Korean AI chip tycoon has launched an appeal against a record R10,7 billion divorce ruling. (IOL)

•  European policy makers are increasingly prioritising AI sovereignty to reduce dependence on United States technology firms. (Whalesbook)

Tencent Releases Hy4

Tencent Announces Hy4 Open Weight Model

Tencent has released Hy4, a new open weight large language model. This model excludes vision capabilities and focuses exclusively on text input. It represents a notable expansion in scale for the company, featuring 770 billion total parameters and 49 billion active parameters. The model supports a 1 million token context window and occupies 1.56TB of data on Hugging Face.

The release follows the July arrival of Hy3, which held 295 billion parameters and 21 billion active parameters with a 256,000 token context window. The move to Hy4 nearly triples the parameter count and quadruples the available context window size.

The practical utility of this increase remains to be seen in real-world application. As the industry moves toward larger parameter sets, developers are increasingly turning to model chat templates to better evaluate performance across different domains. The storage requirement of 1.56TB suggests that deployment will necessitate significant hardware resources for those looking to self-host this particular iteration. (Simon Willison)

Researchers Introduce FreeToken Inference Engine

A team from UC Berkeley and MIT has released FreeToken, an open-source inference engine designed to improve the performance of Mixture-of-Experts models on consumer hardware. The engine utilizes a dynamic scheduling policy and refined weight management to boost decoding speeds for edge applications.

By optimising how models execute on smaller machines, FreeToken aims to make frontier reasoning systems more accessible for self-hosting. The project addresses the historical bottleneck of running high-parameter models on localized hardware configurations. (InfoQ)

Debian Formalises Generative AI Policy

The Debian project has voted to permit the responsible use of generative AI within its ecosystem. This decision establishes a framework for how contributors may integrate automated tooling into the development of their open-source distributions.

The shift acknowledges the growing presence of generative models in modern software engineering workflows. The project aims to balance the utility of these tools against the community standards that govern the production of its software. (Hacker News)

In Brief

•  Representative Gabe Evans has introduced H.R. 10152, known as the Open-Source AI Leadership Act, to the United States Congress. (Quiver Quantitative)

•  Fabiane Nardon presented on architecting data layers for agents using MCP and semantic models to manage token overhead. (InfoQ)

Copyright Disputes

Sony Music And Warner Chappell Sue Anthropic

Sony Music and Warner Chappell have filed a lawsuit against Anthropic in the US District Court for the Northern District of California. The complaint alleges that the artificial intelligence firm used tens of thousands of copyrighted works without authorisation to train its models. The plaintiffs seek damages of up to 150,000 dollars per work.

The labels further claim the company removed identifiable copyright data from these works. They are asking for an additional 25,000 dollars for every instance of such removal. Anthropic has not yet responded to the specific allegations regarding its training data practices.

This action highlights the increasing friction between creators and technology firms. It remains to be seen how the court will interpret the use of intellectual property in the development of large language models. The outcome could set a precedent for how music publishers and AI companies interact in the future. (The Verge)

Australia Bans AI Music From Official Charts

The Australian music industry has officially moved to exclude AI-generated tracks from its national music charts. The decision ensures that only songs created by human artists are eligible for ranking.

The move reflects growing concern among industry leaders regarding the proliferation of algorithmically derived music. By setting this boundary, the industry aims to preserve the value of traditional performance and composition. Questions persist regarding the methods used for verifying the human origin of submitted tracks. (The Business Standard; Scripps News)

Rockstar Games Confirms Stance On GTA 6

Rockstar Games has confirmed that GTA 6 will be released without microtransactions or generative AI. The announcement provides clarity to fans concerned about the potential integration of automated content tools.

The exclusion of generative AI indicates a preference for human-authored content within the studio's latest title. This approach differentiates the release from other industry trends currently favouring the automation of assets. Whether this strategy will influence future titles from the studio remains unknown. (Khel Now; GamesReviews.com)

In Brief

•  Musicians are increasingly investigating online music platforms to identify AI-generated tracks that mimic their work without consent. (The Verge)

•  Nvidia’s DLSS 5 AI filter has leaked to a negative reception from gamers who criticise the visual output as unsettling. (Creative Bloq)

•  The Acer Nitro V 16 AI hardware fails to meet performance expectations when tested with various creative AI software tools. (Creative Bloq)