The Nvidia scandal stopped being about one man's judgment this week. New research says China's most powerful open-weight models were trained on chips Washington banned from export, and the story behind that is bigger than any keynote.
Export-banned Nvidia chips are behind China's most powerful AI models
Will DePue, a former OpenAI researcher now at the nonprofit Unemployment Labs, analyzed the weights of models openly released by DeepSeek, Moonshot AI, MiniMax and Ant Group, and found they still carry the configuration fingerprints of Blackwell GPUs, which are banned from export to China. His conclusion: Chinese frontier labs have widespread access to the very chips export controls were designed to keep from them. TechCrunch notes the finding also sharpens Nvidia's own legal exposure, since corporate liability for export violations may have factored into Jensen Huang's exit this week. Source: https://techcrunch.com/2026/11/13/export-banned-nvidia-chips-are-behind-chinas-most-powerful-ai-models-uh-oh-jensen/
XRAID brings fast declustered RAID straight into XFS
Christoph Hellwig announced the first public release of XRAID, a declustered parity RAID implementation built into XFS. It targets zoned devices like SMR hard drives and claims writes up to 20 times faster than mdraid without mdraid's data integrity problems. Eighteen months in the making, not yet ready for upstream submission, but testers are welcome. Source: https://www.phoronix.com/news/XRAID-XFS-RAID
Anthropic's Glasswing promised to save maintainers. VulnCheck ran the numbers
Patrick Garrity at VulnCheck audited seven months of Anthropic's Project Glasswing, the AI vulnerability-hunting effort launched in April. Of 37,966 findings, only 19.4% were ever made public and just 217 are marked fixed, while duplicates outnumber fixes 7.5 to one and 419 findings have been withdrawn. His verdict: the triage burden Glasswing was supposed to lift is landing on the same open source maintainers it promised to help. The receipts do not reconcile. Source: https://lwn.net/Articles/1107173/
ChatGPT's new Wallet lets its agents do the shopping
OpenAI launched Wallet on Friday, letting ChatGPT's shopping agents pick products and check out on your behalf. It starts with a limited group of US Pro subscribers with an American credit card, and every purchase still needs your Face ID or fingerprint approval. United Airlines blocked this exact behavior on Thursday. OpenAI is betting your wallet that the airlines are wrong. Source: https://techcrunch.com/2026/11/13/chatgpts-new-wallet-lets-ai-agents-shop-and-check-out-for-you/
AMD's open-source firmware hits its production goal
Phoronix confirms AMD's openSIL project, the open-source replacement for the closed AGESA silicon initialization code, has reached production readiness on the new Zen 6 EPYC 9006 "Venice" chips. Coreboot fans have been waiting since 2023 for this. Open firmware on servers is no longer a science project. Source: https://www.phoronix.com/news/AMD-openSIL-Venice
A disgruntled researcher just handed hackers a Windows 0-day
Ars Technica reports a researcher, unhappy with how their report was handled, published a zero-day in Windows that grants SYSTEM rights, proof of concept included, before Microsoft shipped a fix. It is a reminder that disclosure policy arguments have bystanders, and the bystanders this time are every unpatched Windows machine on the internet. Source: https://arstechnica.com/security/2026/11/windows-0-day-dropped-by-disgruntled-researcher-hands-hackers-system-rights/