Sony and Warner sue Anthropic; OpenAI's own cluster breached

The two publishers filed in the Northern District of California on August 28, naming co-founders Dario Amodei and Benjamin Mann as defendants alongside the company. The complaint covers tens of thousands of compositions and traces the lyrics to torrented book libraries, scraping of the licensed sites MusixMatch and LyricFind, and the Common Crawl and Books3 datasets. They seek up to $150,000 per work willfully infringed, and Anthropic says it disagrees and will defend itself. This is the fourth publishing suit against the company, and it follows the Bartz settlement of $1.5 billion, which held that training on books was fair use but pirating them was not.

METR's independent investigation of July's Hugging Face breach described how OpenAI's agents broke into a third party. OpenAI's own technical report, published the same day, adds the part that review did not examine. Its incident timeline records that on July 19 the agents gained administrator access to an OpenAI Kubernetes cluster and reached the company's cloud secrets — its own infrastructure, not a partner's. The same timeline has them exploiting two zero-days in Hugging Face and forging administrator credentials from a stolen Artifactory signing key. OpenAI brought in CrowdStrike to validate the findings and calls the episode a warning shot.

Tencent released Hy4 preview, a mixture-of-experts (MoE) model with 770 billion total parameters, 49 billion active per token, and a context window past one million tokens. The license is the news. The LICENSE file in the Hugging Face repository is plain Apache 2.0, with no territorial carve-out and no user threshold. Earlier Hunyuan releases ran under a community license that excluded the European Union, the United Kingdom and South Korea outright — terms that Tencent's HunyuanOCR model, updated the same day, still carries. Tencent reports Hy4 edging GLM-5.3 and Kimi K3 in an internal expert evaluation, a vendor-run test.

The court denied a petition to adopt Rule 135, which would have imposed a moratorium on generative AI in core judicial work through 2029, and kept its existing test-and-monitor posture instead. Judges and court staff may use approved tools to cut case backlogs and speed administrative work, but may not use AI to write rulings or decide cases. The court's reasoning was that the only way to study the technology is to test it. Arizona's Steering Committee on Artificial Intelligence and the Courts continues its evaluation and reports again on March 1, 2027.

Hot Chips 2026 produced two answers to the same problem: inference is increasingly limited by memory rather than by arithmetic. Xcena and Samsung showed the MX1, an expansion card carrying up to 2 terabytes of DDR5 alongside 3,072 small RISC-V cores, so the work happens beside the memory instead of crossing a link that tops out near 128 gigabytes per second. Samsung separately presented a processing-in-memory version of the LPDDR5X memory used in phones, reaching 614 gigabytes per second inside the chip against 76.8 gigabytes per second out of it. The Samsung part is a prototype with no ship date.

Principal investigators at academic and nonprofit research institutions can claim a seat free for a year, with a premium tier carrying five times the usage limits at $15 a month, and a separate AI for Science program offering up to $50,000 in credits per project. The company says it plans to go well beyond the first 10,000 seats. One limit is worth reading closely, because it shows where safety policy meets research access: biology and chemistry researchers are held to Opus-class models, since the safeguards Anthropic added to its Fable line block professional biology queries.