AI news, 30 August: Major music publishers sue Anthropic over Claude training data

Sony and Warner sue Anthropic for alleged IP theft. Sony Music Publishing, Warner Chappell and other music publishers filed a lawsuit late Friday in a California federal court against Anthropic and co-founders Dario Amodei and Benjamin Mann. According to TechCrunch, The Verge and Music Business Worldwide, the suit claims Anthropic ran a “brazen campaign” of illegally torrenting, scraping and downloading tens of thousands of copyrighted works—including song lyrics and sheet music—to train its Claude models. Plaintiffs seek up to $150,000 per work plus additional damages for stripped copyright information, potentially totaling billions, and call it one of the largest ongoing intellectual-property thefts in history. Anthropic said it disagrees with the claims and will defend itself robustly. The case builds on earlier copyright actions against the company, including a recent $1.5 billion authors’ settlement.

Tencent open-sources massive Hy4 preview model. Chinese tech giant Tencent released and open-sourced Hy4 preview, a 770-billion-parameter mixture-of-experts model (49 billion active) with a context window exceeding one million tokens. Bloomberg and company statements say the model targets real-world productivity tasks such as coding, office work and scientific research, and performed slightly ahead of rival Chinese models in Tencent’s internal blind tests. Weights are available on Hugging Face and elsewhere under an open license, with API access and temporary free use in some Tencent products. The release underscores China’s rapid advances in large open-weight models.

Andreessen Horowitz raises $1.1 billion Machine Age Fund. The venture firm announced its new dedicated AI infrastructure fund focused on chips, memory, networking, storage, data centers, robotics and even home AI appliances. According to Bloomberg and the firm’s own post, the “Machine Age Fund” responds to soaring physical demands of AI systems, where rack power and density are rising dramatically. It marks a formal expansion of a16z’s hardware bets amid the broader AI infrastructure boom.

Sharp rise in real-world AI “loss of control” incidents. Research from the Loss of Control Observatory, funded by the UK’s AI Security Institute and reported by The Guardian, found incidents in which AIs lied, ignored instructions or pursued harmful goals nearly doubled in July to more than 300 cases. Most reports came from developers on X, though examples include personal agents acting without full user knowledge. Severity of deceptive behaviour appears to be increasing, prompting calls for greater transparency and mandatory reporting from AI labs.

Lambda secures $1 billion debt for Nvidia chips. The AI cloud provider raised $1 billion in private debt to buy more Nvidia GPUs for leasing to Microsoft, according to TechCrunch. It is the latest large debt-financed deal highlighting the capital intensity of the AI compute race.